약관 및 정책

개인정보 처리방침

서비스
Human.ai (휴먼ai)
시행일
2026년 11월 1일

크레이빙(CRAVING)(이하 ‘운영자’)은(는) Human.ai(휴먼ai, 이하 ‘서비스’)를 운영하면서 「개인정보 보호법」 제30조에 따라 이용자의 개인정보를 보호하고, 개인정보와 관련한 고충을 신속하고 원활하게 처리할 수 있도록 다음과 같이 개인정보 처리방침을 수립하여 공개합니다.

이 처리방침에서 ‘성향 리포트’는 연애 성향 리포트, 관계 안정감 리포트, 대인관계 스타일 리포트 세 가지를, ‘셀프체크’는 마음 컨디션 체크, 집중 스타일 리포트 두 가지를, ‘인스타그램 리포트’는 인스타그램 피드 리딩 기능을 가리킵니다. 스토어에서 내려받는 모바일 앱에는 셀프체크가 들어 있지 않으며, 셀프체크에 관한 내용은 셀프체크가 제공되는 곳에만 적용됩니다.

법률 검토 전 초안

이 문서는 변호사의 검토를 마치기 전의 초안이며 확정본이 아닙니다. 점선으로 표시한 대괄호 부분은 운영자가 채워야 하는 정보이고, ‘검토 메모’는 검토하는 사람을 위한 안내입니다.

  • Human.ai에는 회원가입과 로그인이 없습니다. 운영자는 이용자의 응답이나 리포트를 저장하는 데이터베이스를 운영하지 않습니다.
  • 서비스는 성인만 이용할 수 있습니다(대한민국 기준 만 19세 이상). 처음 시작할 때 생년월일을 묻지만 나이를 계산하는 데만 쓰고 저장하거나 전송하지 않으며, 확인을 마쳤다는 사실만 기기에 남습니다.
  • 문항 응답, 과제 기록, 그림, 완성된 리포트는 이용자의 기기에만 저장됩니다.
  • 셀프체크는 기기 안에서만 계산하며, 이와 관련한 어떤 내용도 서버로 전송하지 않습니다.
  • AI 해석은 이용자가 동의하고 요청한 경우에만 만들어집니다. 이때 필요한 내용만 운영자의 서버를 거쳐 미국의 AI 서비스로 전송하며, 운영자의 서버는 사본을 보관하지 않습니다. AI 서비스 제공자가 전달받은 내용을 보유하는 기간은 개인정보의 국외 이전에 적습니다.
  • 앱에서 인스타그램 리포트의 심층 리딩을 구매하면 결제는 앱 마켓이 처리하며, 운영자는 이름, 이메일 주소, 결제 수단의 정보를 받지 않습니다. 구매가 실제인지 확인한 뒤에는 거래 번호의 해시값과 사용 기록만 400일 동안 보관합니다.
  • 광고나 분석을 위한 추적 도구를 사용하지 않습니다. 쿠키는 언어 선택을 기억하는 것 하나와, Instagram 로그인 연결을 진행하는 동안에만 두는 보안용 하나뿐입니다.

1.개인정보의 처리 목적

운영자는 다음의 목적을 위하여 필요한 최소한의 개인정보를 처리합니다. 처리하는 개인정보는 다음의 목적 이외의 용도로 이용하지 않으며, 이용 목적이 바뀌는 경우에는 관련 법령에 따라 별도의 동의를 받는 등 필요한 조치를 합니다.

  • AI 해석의 제공: 성향 리포트에 대하여 이용자가 요청한 AI 해석을 작성하여 제공합니다.
  • 인스타그램 리포트의 제공: 이용자가 직접 고른 본인 계정의 게시물을 읽고, 피드가 보여 주는 모습을 정리한 리포트를 작성하여 제공합니다. 무료인 기본 리딩과 첫인상 리딩, 앱에서 구매하는 유료 심층 리딩이 있습니다.
  • 구매의 확인: 앱에서 구매한 심층 리딩의 결제가 실제로 이루어졌는지 확인하고, 한 번의 구매가 여러 번의 리딩에 쓰이지 않도록 합니다.
  • 서비스의 안정적인 운영과 보안: 접속 기록과 AI 요청의 처리 기록(요청 내용은 제외)을 관리하고, 과도한 요청과 하루 사용량을 제한하며, 장애와 부정 이용에 대응합니다.
  • 문의와 권리 행사 요청의 처리: 이용자나 정보주체가 이메일이나 전화로 보낸 문의(환불 요청 포함)와 열람, 정정·삭제, 처리정지 요청을 확인하고 회신합니다.

2.처리하는 개인정보의 항목과 법적 근거

동의를 받아 처리하는 개인정보

성향 리포트의 AI 해석

처리 항목

운영자의 서버가 받는 것: 채점 결과(척도별 점수와 그 범위, 백분위·구간, 답한 문항 수, 유형, 응답 품질 지표, 검사를 마친 시각), 리포트의 그림과 표를 다시 그리는 데 쓰는 파생 수치(리포트에 따라 좌표와 세부 점수, 가까운 유형의 코드, 검사 형식과 비교 기준의 구분, 답한 문항 수, 채점하지 못한 척도와 답하지 않은 문항의 번호, 같은 답이 가장 길게 이어진 횟수와 가장 많이 고른 답의 횟수, 서로 반대되는 문항 쌍에 어긋나게 답한 횟수, 문항당 응답 시간의 중앙값), 리포트의 종류와 버전, 표시 언어. 문항마다 고른 응답은 전송하지 않습니다.

이 가운데 AI 서비스 제공자에게 전달하는 것: 리포트의 종류와 버전, 표시 언어, 척도의 이름과 점수(범위 포함), 백분위·구간, 답한 문항 수, 유형의 이름과 한 줄 설명, 응답 품질에 관한 안내 문장. 검사를 마친 시각과 위의 파생 수치는 서버가 쓰지 않으며 전달하지도 않습니다. 모든 이용자에게 같은 검사 문항의 문구가 점수의 뜻을 설명하는 자료로 함께 주어집니다.

처리 목적

AI 해석 문장의 작성과 제공

법적 근거

정보주체의 동의(「개인정보 보호법」 제15조제1항제1호)

처리 시점

이용자가 AI 해석에 동의하고 리포트에서 AI 해석을 요청한 때

동의의 기록과 철회

동의한 사실과 그 시각은 이용자의 기기에만 기록되며 운영자에게 전송되지 않습니다. 동의는 성인 확인을 마친 기기에서만 받습니다. 설정에서 언제든지 철회할 수 있고, 철회한 뒤로는 아무것도 전송하지 않습니다.

보유 기간

운영자의 서버에 저장하지 않습니다. 요청을 처리하는 동안 메모리에서만 다루며, 결과를 기기로 돌려준 뒤에는 남기지 않습니다.

운영자의 위탁을 받은 AI 서비스 제공자의 보유·이용 기간: 받거나 생성한 날부터 30일 이내에 삭제합니다. 다만 이용정책 위반으로 분류된 입력과 출력은 최대 2년, 안전 분류 점수는 최대 7년까지 보관될 수 있습니다.

동의를 거부할 권리

동의하지 않을 수 있습니다. 이 경우에도 척도 점수에 따른 기본 해석이 담긴 리포트를 그대로 이용할 수 있습니다.

인스타그램 리포트

처리 항목

게시물을 분류하는 단계: 이용자가 직접 고른 본인 계정 게시물의 축소한 이미지(긴 변 768픽셀 이하, 촬영 정보 등 메타데이터 제거)와 글(캡션), 게시물에 순서대로 붙인 번호, 리포트의 언어, 자료의 종류(게시물의 사진인지 화면 캡처인지)와 자료를 가져온 방법(화면 캡처, 내보내기 파일, Instagram 로그인 연결 가운데 하나). 내보내기 파일로 넘겨주신 여러 장으로 된 게시물은 표지 다음 사진을 4장까지(긴 변 512픽셀 이하, 메타데이터 제거) 함께 보내고, 사진마다 그 게시물의 번호와 순서를 붙입니다. 글 속에 있는 다른 계정의 이름(@)은 가린 뒤에 보냅니다. 자료를 가져온 방법은 서버가 이름을 가리는 데에만 쓰며 AI 서비스 제공자에게 전달하지 않습니다.

해석하는 단계: 사진은 다시 보내지 않고 글자로 된 근거 자료만 보냅니다. 게시물마다의 번호, 게시 날짜(내보내기 파일이나 Instagram 로그인 연결로 넘겨준 경우), 게시물 종류, AI가 분류 단계에서 붙인 중립적인 한 줄 묘사와 분류값(사진 속에 보이는 글자를 옮겨 적은 것 포함), 글의 앞부분(300자까지), 그리고 기기에서 계산한 집계값(게시물 수와 비율, 여섯 축의 점수와 피드 유형, 올리는 리듬과 색의 통계)입니다.

첫인상 리딩(보낼 수 있는 게시물이 12개보다 적을 때, 또는 분류한 뒤 읽을 수 있는 게시물이 그보다 적을 때 제공하는 무료 리딩): 분류와 해석을 나누지 않고 한 번의 요청으로 보냅니다. 이용자가 고른 본인 계정의 이미지 11장까지를 분류 단계와 같은 방법으로 축소하고 메타데이터를 제거하여 보내며, 이미지마다 붙인 번호, 글(캡션)이 있으면 그 글, 리포트의 언어, 자료를 가져온 방법을 함께 보냅니다. 글 속에 있는 다른 계정의 이름(@)은 가린 뒤에 보내고, 자료를 가져온 방법은 서버가 이름을 가리는 데에만 씁니다. 게시 날짜와 집계값은 보내지 않습니다. 분류한 뒤에 제공하는 경우에는 분류 단계에서 어떤 해석의 근거로도 쓰지 않기로 한 게시물을 보내지 않습니다.

화면 캡처를 고른 경우 캡처에 찍혀 있는 내용(계정 이름, 프로필 소개글, 팔로워·좋아요 수, 다른 사람의 댓글 등)은 사진의 일부로 그대로 전송됩니다. 그 밖에는 계정 이름, 프로필 소개글, 좋아요·댓글·팔로워 수를 서버로 보내지 않습니다. 다만 Instagram 로그인으로 연결한 경우에는 서버가 본인 계정에서 이 가운데 일부를 직접 읽으며(아래 표), 그때에도 계정 이름, 소개글, 좋아요·댓글 수는 AI 서비스 제공자에게 전달하지 않습니다. 이용자가 제외한 게시물은 사진과 글을 보내지 않으며, 게시 날짜만 올리는 리듬의 집계에 반영됩니다.

자료를 넘겨주는 방법

이용자가 고른 본인 계정의 화면 캡처나 사진(피드 게시물, 여러 장 게시물의 한 장, 스토리·하이라이트, 프로필 화면의 캡처), 이용자가 인스타그램에서 내려받아 기기에서 읽어 들인 ‘내 정보 내보내기’ 파일, 또는(제공되는 경우) 아래의 Instagram 로그인 연결.

‘내 정보 내보내기’ 파일은 기기 안에서 읽으며, 파일 자체를 서버로 전송하지 않습니다.

민감정보

게시물에는 건강, 신념 등 민감정보가 드러나는 내용이 포함될 수 있습니다. 이 때문에 다른 동의와 구분하여 별도의 동의를 받습니다.

처리 목적

인스타그램 리포트의 작성과 제공

법적 근거

정보주체의 동의(「개인정보 보호법」 제15조제1항제1호)

민감정보 처리에 대한 별도의 동의(같은 법 제23조제1항제1호)

국외 이전에 대한 별도의 동의(같은 법 제28조의8제1항제1호)

동의를 받는 방법

리딩을 시작할 때마다 네 가지를 각각 따로 확인받습니다. 제출하는 자료가 본인 계정의 것이라는 확인, 개인정보 수집·이용 동의, 민감정보 처리 동의, 국외 이전 동의입니다. 이 동의는 그 리딩에만 적용되고 어디에도 저장하지 않습니다. 첫인상 리딩에도 같은 네 가지를 확인받습니다. 동의는 보낼 사진과 글을 보여 주는 마지막 확인 화면에서 받으며(Instagram 로그인으로 연결하는 경우에는 연결하기 전에 받습니다), 동의한 뒤에도 그 화면에서 시작 버튼을 누르기 전에는 아무것도 전송하지 않으며, 전송이 시작된 뒤에 취소하면 그 뒤로는 전송하지 않습니다. AI 서비스를 쓸 수 없을 때 고를 수 있는, 내보내기 파일을 기기에서 세기만 하는 리포트는 아무것도 전송하지 않으므로 본인 계정의 자료라는 확인만 받습니다.

보유 기간

운영자의 서버에 저장하지 않습니다. 메모리에서만 처리하고, 결과는 이용자의 기기로 돌려보냅니다.

운영자의 위탁을 받은 AI 서비스 제공자의 보유·이용 기간: 받거나 생성한 날부터 30일 이내에 삭제합니다. 다만 이용정책 위반으로 분류된 입력과 출력은 최대 2년, 안전 분류 점수는 최대 7년까지 보관될 수 있습니다.

같은 게시물을 다시 보내지 않도록, AI가 게시물마다 붙인 분류값과 한 줄 묘사(화면 캡처에서 읽은 글 포함)를 이용자의 기기에 7일 동안 보관합니다. 사진은 여기에 저장하지 않습니다.

동의를 거부할 권리

동의하지 않을 수 있습니다. 이 경우 인스타그램 리포트를 이용할 수 없으며, 다른 리포트의 이용에는 제한이 없습니다.

Instagram 로그인 연결(선택 기능, 웹에서만 제공)

적용되는 경우

프로페셔널 계정을 가진 이용자가 본인 계정을 직접 연결하는 경우에만 해당합니다. 운영자가 이 기능을 제공하지 않는 동안에는 해당 사항이 없습니다.

요청하는 권한

기본 읽기 권한(instagram_business_basic) 하나만 요청합니다. 게시, 메시지, 댓글, 인사이트에 관한 권한은 요청하지 않습니다.

처리 항목

Instagram이 발급한 일회용 인증 코드와 단기 접근 토큰, 계정 이름과 표시 이름, 프로필 소개글, 최근 게시물(최근 36개월 이내, 최대 60개)의 글(캡션), 게시 일시, 게시물 종류와 앨범의 사진 수, 표지 이미지, 좋아요 수와 댓글 수

처리 방식

운영자의 서버가 이용자의 토큰으로 Instagram에서 위 내용을 읽고, 표지 이미지를 축소하여 이용자의 기기로 전달합니다. 계정 이름으로 조회하는 것이 아니며, 토큰의 주인인 본인 계정만 읽습니다.

서버는 토큰, 게시물, 이미지를 저장하지 않습니다. 토큰은 암호화하여 이용자의 브라우저에만 맡기며, 30분이 지나면 서버가 받아들이지 않습니다. 브라우저는 이를 열려 있는 페이지의 메모리에만 두고 저장 공간이나 쿠키에 기록하지 않으므로, 페이지를 새로 고치거나 닫으면 사라집니다.

Instagram에서 서비스로 돌아오는 주소에는 위의 일회용 인증 코드가 들어 있으며, 이 주소는 다른 요청과 마찬가지로 호스팅 사업자의 접속 기록에 남을 수 있습니다. 인증 코드는 한 번만 쓸 수 있고, 서버는 이를 받는 즉시 토큰으로 바꿉니다.

운영자는 Instagram 비밀번호를 받지 않습니다. 로그인과 승인은 Instagram의 화면에서 이루어집니다.

처리 목적

인스타그램 리포트에 쓸 본인 계정의 게시물을 이용자의 기기로 가져오기

법적 근거

정보주체의 동의(「개인정보 보호법」 제15조제1항제1호). 이용자가 직접 연결을 시작하고 Instagram의 동의 화면에서 승인한 경우에만 처리합니다.

보유 기간

운영자의 서버에 저장하지 않습니다.

동의를 거부할 권리

연결하지 않을 수 있습니다. 이 경우에도 화면 캡처나 사진, 또는 ‘내 정보 내보내기’ 파일로 인스타그램 리포트를 이용할 수 있습니다.

연결 해제

서비스에서 연결을 해제하면 브라우저가 가지고 있던 연결 정보를 바로 버립니다. 리딩을 마치거나 그 화면을 떠날 때에도 자동으로 버립니다. 서버에는 지울 토큰이 없습니다. Instagram이 발급한 토큰은 수명이 1시간이며, 운영자는 이를 수명이 더 긴 토큰으로 바꾸거나 연장하지 않습니다.

삭제를 요청하는 방법

운영자의 서버에는 Instagram에서 읽어 온 내용이 남아 있지 않습니다. 기기에 저장된 리포트는 설정에서 직접 지울 수 있으며, 그 밖의 요청은 contact@craving.win(으)로 할 수 있습니다.

검토 메모

AI 서비스로 전송하는 항목은 요청 형식(src/lib/contracts.ts)과 서버 코드(src/server/ai)를 읽고 적었으며, 일부는 자동 검사(legal.test.ts)가 지켜봅니다. 남은 검토 사항은 다음과 같습니다. 화면 캡처에는 다른 사람의 계정 이름과 댓글이 찍혀 있을 수 있고 사진의 일부로 전송되므로, 정보주체가 아닌 사람의 개인정보 처리로 다루어야 하는지 판단이 필요합니다. 첫인상 리딩은 스토리, 하이라이트, 프로필 화면의 캡처도 받으므로 이 판단이 더 넓은 자료에 적용됩니다. AI 서비스 제공자의 보유 기간(받거나 생성한 날부터 30일 이내에 삭제합니다. 다만 이용정책 위반으로 분류된 입력과 출력은 최대 2년, 안전 분류 점수는 최대 7년까지 보관될 수 있습니다.)과 연락처는 Anthropic이 공개한 안내 페이지에서 옮겨 적었으며(2026년 10월 20일 확인), 동의 화면도 같은 숫자를 적고 있습니다. 운영자가 동의한 상용 약관과 데이터 처리 부속 계약에 비추어 이 내용이 맞는지 확인하고, 다르면 두 곳이 같은 내용을 말하도록 함께 고쳐야 합니다. 동의 화면은 여기에 더하여 법령상 보관 의무가 있으면 그 기간 동안 보관될 수 있다는 점과 AI 모델의 학습에 이용되지 않는다는 점을 적고 있으므로, 이 두 가지도 계약 조건으로 확인하여 처리방침에 함께 적을지 정하여야 합니다. 성향 리포트의 동의 화면은 이전되는 항목에 검사를 마친 시각을 넣고 있으나 서버는 이를 AI 서비스 제공자에게 전달하지 않습니다. 동의의 기록은 이용자의 기기에만 있고(인스타그램 리포트의 동의는 어디에도 남지 않습니다) 운영자는 동의를 받았다는 증빙을 보유하지 않으므로, 이 구조가 동의의 입증 책임에 비추어 받아들여질 수 있는지 검토가 필요합니다. 연애 성향 리포트의 채점 결과에는 현재의 연애 관계에 관한 점수(친밀감, 열정, 헌신)가 포함될 수 있으므로, 성향 리포트를 민감정보가 아닌 것으로 본 분류를 확정할 때 함께 검토하여야 합니다.

검토 메모

Instagram 로그인 연결은 Meta의 앱 검수와 플랫폼 약관의 적용을 받으므로, 이 처리방침이 플랫폼 약관 제4조가 요구하는 내용(처리하는 데이터, 처리 방법과 목적, 삭제 요청 방법)을 충족하는지 확인이 필요합니다. 연결 화면과 서버 코드는 만들어져 있으나 실제 Meta 앱으로 동작시켜 확인한 적은 없습니다. 글(캡션)이 이 연결로 실제로 제공되는지도 확인되지 않았으므로, 운영자가 이 기능을 켜기 전에 위 항목을 실제 응답과 대조하여야 합니다.

동의 없이 처리하는 개인정보

다음 정보는 서비스를 제공하고 안전하게 운영하는 데 필요하여 동의 없이 처리합니다.

접속 기록과 서버의 처리 기록

처리 항목

IP 주소, 접속 일시, 요청한 주소, 브라우저와 기기 정보 등 호스팅 사업자의 서버에 통상적으로 남는 접속 기록.

AI 요청의 처리 기록: 처리 단계의 이름, 결과(성공, 거절, 잘림, 형식 오류, 실패)와 오류의 분류, 사용한 모델의 이름과 응답이 끝난 사유, 걸린 시간, 입력·출력·캐시 토큰 수, 추론이 이루어진 지역, 처리한 게시물·이미지·해석 문장·질문·리포트 장의 개수. 이 기록에는 숫자, 참·거짓, 짧은 식별 낱말만 적을 수 있게 되어 있어 점수, 사진, 글, 생성된 문장 같은 요청과 응답의 내용은 남지 않습니다.

서버에서 예상하지 못한 오류가 난 경우에는 오류의 종류(이름)만 남습니다. 오류 메시지와 발생 위치는 요청 내용을 담을 수 있어 남기지 않습니다. Instagram 로그인 연결에서 오류가 난 경우에는 오류의 분류와 Instagram이 돌려준 상태 코드와 오류 코드가 남습니다.

처리 목적

서비스의 제공, 장애 대응, 보안

법적 근거

「개인정보 보호법」 제15조제1항제4호(이용자가 요청한 서비스의 제공), 같은 항 제6호(서비스의 안전한 운영을 위한 정당한 이익)

보유 기간

30일

요청 횟수 제한을 위한 값

처리 항목

IP 주소를 비밀값과 함께 일방향 해시 처리한 값과 요청 횟수. 이 용도로는 IP 주소 원문을 저장하지 않습니다. 이용자를 구분하지 않고 서비스 전체의 하루 AI 요청 수를 세는 값도 함께 둡니다.

처리 목적

AI 해석, 인스타그램 리포트, Instagram 로그인 연결 요청이 짧은 시간에 지나치게 몰리는 것을 막고, 하루에 처리하는 AI 요청의 총량에 상한을 두기 위한 요청 횟수 제한

법적 근거

「개인정보 보호법」 제15조제1항제4호(이용자가 요청한 서비스의 제공), 같은 항 제6호(서비스의 안전한 운영을 위한 정당한 이익)

보유 기간

만료 시간이 설정되어 있어, 처음 기록된 때부터 제한 구간의 길이가 지나면 자동으로 삭제됩니다. 제한 구간은 AI 해석과 인스타그램 리포트 요청이 10분과 하루(인스타그램 리포트의 하루 제한은 기본 리딩과 첫인상 리딩에만 적용), Instagram 로그인 연결 요청이 10분, 제공 기능을 확인하는 요청이 1분입니다.

심층 리딩 구매의 확인

적용되는 경우

앱에서 인스타그램 리포트의 심층 리딩을 구매한 경우에만 해당합니다. 무료인 기본 리딩에는 해당하지 않습니다.

처리 항목

앱이 서버로 보내는 것: 구매한 앱 마켓이 그 구매에 대하여 앱에 건네준 정보뿐입니다. 앱 마켓에 따라 둘 가운데 하나입니다. 하나는 앱 마켓이 서명한 거래 기록으로, 앱 마켓이 발급한 그대로 전부 보냅니다. 여기에는 거래 번호, 앱과 상품, 구매 시각, 테스트 결제인지 실제 결제인지와 함께 최초 거래 번호와 앱 거래 번호, 스토어 국가, 가격과 통화, 기기를 확인하는 값, 그 밖에 앱 마켓이 기록에 넣은 값이 들어 있습니다. 다른 하나는 상품 ID와 구매 토큰입니다. 심층 리딩의 게시물 분류 요청과 해석 요청에 함께 보냅니다.

서명된 거래 기록은 서버가 메모리에서 그 서명을 직접 검증하며, 이를 위해 앱 마켓에 따로 연락하지 않습니다. 확인에는 그 가운데 거래 번호, 앱과 상품, 상품의 종류, 결제 환경, 구매 시각과 서명 시각, 환불 여부만 쓰고 나머지 값은 쓰지 않습니다. 구매 토큰을 받은 경우에는 서버가 이를 앱의 식별자, 상품 ID와 함께 그 앱 마켓의 결제 시스템에 보내 주문 번호와 구매 상태를 확인하고, 리딩을 전달한 뒤 그 구매가 사용되었음을 알립니다.

서버가 보관하는 것: 앱 마켓의 구분과 거래 번호를 함께 일방향 해시 처리한 값과, 그 구매로 받은 리딩, 리딩 시도, 게시물 분류 요청의 횟수와 처음 리딩을 받은 시각.

받지 않는 정보

이름, 이메일 주소, 카드 번호 같은 결제 수단의 정보, 앱 마켓 계정 정보는 받지 않습니다. 결제는 앱 마켓이 직접 처리합니다.

처리 목적

구매가 실제로 이루어졌는지 확인하고, 한 번의 구매가 여러 번의 리딩에 쓰이지 않도록 하기 위함

법적 근거

「개인정보 보호법」 제15조제1항제4호(이용자와 체결한 계약의 이행), 같은 항 제6호(한 번의 구매가 여러 번 쓰이지 않도록 하기 위한 정당한 이익)

보유 기간

거래 번호의 해시값과 사용 기록: 요청 횟수 제한에 쓰는 저장소(Upstash, Inc.)에 400일 동안 보관한 뒤 자동으로 삭제합니다. 리딩이 진행되는 동안에는 같은 구매로 리딩이 겹치지 않도록 잠금 표시를 최대 5분 동안 둡니다.

앱이 보낸 거래 기록과 구매 토큰: 저장하지 않습니다. 확인하는 동안 서버의 메모리에서만 다룹니다.

문의와 권리 행사 요청

처리 항목

이메일로 문의한 경우 보낸 사람의 이메일 주소, 전화(010-9027-0729)로 문의한 경우 건 사람의 전화번호, 그리고 문의 내용

환불이나 구매에 관한 문의인 경우 알려 준 구매 정보(구매한 앱 마켓과 주문 번호). 주문 번호는 ‘심층 리딩 구매의 확인’에 적은 구매 사용 기록과 대조할 수 있으므로, 이를 알려 주면 운영자가 해당 구매의 사용 기록을 찾을 수 있습니다.

처리 목적

문의와 환불 요청에 대한 회신, 권리 행사 요청의 확인과 처리

법적 근거

「개인정보 보호법」 제15조제1항제4호(정보주체의 요청에 따른 조치의 이행)

보유 기간

3년

검토 메모

접속 기록, AI 요청의 처리 기록, 요청 제한 값, 문의 기록에 적은 법적 근거는 초안 작성자의 제안입니다. 리서치 문서에 직접적인 근거가 없으므로 확정 전에 검토가 필요합니다. 호스팅 사업자(Vercel Inc.)가 런타임 로그(AI 요청의 처리 기록 포함)를 보관하는 기간은 요금제에 따라 1시간(Hobby), 1일(Pro), 30일(Pro에서 Observability Plus를 쓰는 경우)로 다릅니다. 운영자의 팀은 Pro이고 Observability Plus가 켜져 있어(2026년 10월 27일 확인) 접속 기록의 보유 기간을 30일로 적었으므로, Observability Plus를 끄거나 이 프로젝트를 제외하면 이 값을 1일로 고쳐야 합니다. 전화로 받은 문의를 녹음하거나 따로 기록한다면 그 항목과 보유 기간도 문의 기록에 추가하여야 합니다. 해시에 섞는 비밀값(APP_SECRET)이 운영 환경에 설정되어 있지 않으면 해시값에서 IP 주소를 쉽게 되찾을 수 있으므로, 설정 여부를 확인하여야 합니다. 예상하지 못한 오류는 오류의 이름만 기록하도록 되어 있으므로, 운영 환경의 로그에 그 밖의 내용이 남지 않는지 확인하여야 합니다. 구매의 확인에 적은 근거(계약의 이행과 정당한 이익)와, 거래 기록과 구매 토큰을 저장하지 않는다는 문장은 결제 코드(src/server/billing)를 읽고 적었으나 운영 환경에서 실제 구매로 확인한 적은 없으므로, 첫 판매 뒤 서버 로그와 저장소를 직접 확인하여야 합니다.

수집하지 않는 정보

  • Human.ai에는 회원가입과 로그인이 없으므로 가입을 위한 이름, 아이디, 비밀번호, 전화번호를 수집하지 않습니다.
  • 결제는 앱 마켓이 직접 처리하므로 이름, 이메일 주소, 카드 번호 같은 결제 수단의 정보, 앱 마켓 계정 정보를 받지 않습니다. 구매를 확인하는 데 받는 정보는 위의 ‘심층 리딩 구매의 확인’에 적은 것뿐입니다.
  • 성향 리포트에서 문항마다 고른 응답은 서버로 전송하지 않습니다.
  • 셀프체크는 기기 안에서만 계산하며, 응답, 과제 기록, 그림, 점수, 결과 어느 것도 서버로 전송하지 않습니다.
  • 성인 확인에 입력한 생년월일은 기기 안에서 나이를 계산하는 데만 쓰고, 저장하지도 서버로 전송하지도 않습니다.
  • 다른 사람의 인스타그램 계정을 계정 이름으로 조회하지 않습니다. 게시물을 자동으로 수집(스크래핑)하지 않으며, 그렇게 수집된 데이터를 구매하지도 않습니다.
  • ‘내 정보 내보내기’ 파일을 읽을 때 다이렉트 메시지, 연락처, 로그인·기기 기록, 검색·광고 기록이 든 폴더는 열지 않으며, 게시물에 붙은 위치 정보도 읽지 않습니다. 글 속에 있는 다른 계정의 이름(@)은 번호로 바꾸어 읽습니다.
  • 위치 정보를 요청하지 않습니다. 도움받을 수 있는 연락처는 기기에 설정된 시간대를 기기 안에서 읽어 지금 있는 곳에 맞는 것을 골라 보여 주며, 그 시간대는 서버로 전송하지 않습니다.

3.이용자의 기기에만 저장되는 정보

다음 정보는 이용자의 기기에만 저장됩니다. 운영자의 서버에는 사본이 없으며, 운영자는 이 정보에 접근할 수 없습니다.

  • 문항 응답과 진행 중인 검사의 상태
  • 주의력 과제의 수행 기록(반응 시간과 정확도, 입력 방식과 화면 크기 같은 기기 정보)과 그림
  • 완성된 리포트. AI 해석을 받은 경우 그 해석이 포함되고, 인스타그램 리포트에는 근거로 인용된 게시물이나 이미지의 작은 미리보기 이미지와 글 일부, 그리고 내보내기 파일이나 Instagram 로그인 연결에서 읽은 경우 계정 이름이 포함됩니다.
  • 성인 확인을 마쳤다는 사실, 기준 연령에 미치지 못하는 생년월일이 입력된 경우 다시 확인할 수 있게 되는 시각, AI 해석 동의 여부와 동의한 시각, 화면 테마와 진동 설정, 앱을 처음 열 때 나오는 소개를 이미 봤는지 여부. 생년월일 자체는 저장하지 않습니다.
  • 인스타그램 리포트를 만들 때 AI가 게시물마다 붙인 분류값과 한 줄 묘사. 같은 게시물을 다시 보내지 않기 위한 것으로 7일 동안 보관하며, 사진은 저장하지 않습니다. 읽기를 도중에 멈추거나 읽는 화면을 나가면 그 읽기에서 받은 분류값은 함께 지웁니다.
  • 모바일 앱에서 심층 리딩을 구매한 경우의 구매 정보. 리딩을 진행 중인 구매의 증빙(앱 마켓이 서명한 거래 기록이나 구매 토큰)과 거래 번호는, 앱을 닫았다가 다시 열어도 그 구매로 리딩을 이어 가고 결과를 받지 못했을 때 다시 요청할 수 있도록 리포트가 도착할 때까지 보관하며, 리포트가 전달되거나 그 구매를 닫거나 따로 두면 지웁니다. 서버가 받아들이지 않았거나 리포트가 전달되지 않은 채 횟수를 다 쓴 구매의 거래 번호는, 그 구매를 다시 쓰지 않도록 가장 최근의 것 50개까지 보관합니다. 이름, 이메일 주소, 결제 수단의 정보는 들어 있지 않습니다.

저장되는 곳은 기기의 브라우저 또는 앱 저장 공간(IndexedDB)입니다. 화면 테마와, 이 기기에 저장된 리포트나 진행 중인 검사가 있는지를 나타내는 표시 하나, Instagram 내보내기 파일을 요청하러 간 시각은 브라우저의 로컬 저장소에 둡니다. 심층 리딩의 구매 정보는 모바일 앱 안에서만 앱의 기본 설정 저장 공간에 두며, 웹에서는 저장하지 않습니다. 다른 기기나 브라우저에서는 보이지 않으며, 앱을 삭제하거나 브라우저에서 이 사이트의 데이터를 지우면 함께 삭제됩니다.

설정의 ‘내 데이터 내보내기’로 리포트, 진행 중인 검사, 설정의 사본을 받을 수 있고, ‘이 기기의 모든 데이터 삭제’로 리포트, 진행 중인 검사, 성인 확인과 동의, 설정, 화면 테마와 언어 선택을 한 번에 지울 수 있습니다. 게시물 분류값의 임시 보관분과, 저장된 리포트나 진행 중인 검사가 있는지를 나타내는 표시, Instagram 내보내기를 요청한 시각도 그 자리에서 함께 지워집니다. 두 가지는 다르게 처리됩니다. 기준 연령에 미치지 못하는 생년월일이 입력되어 다시 확인할 수 있게 되는 시각이 아직 지나지 않았다면, 그 시각은 지날 때까지 남습니다. 심층 리딩의 구매 정보는 이 기능으로 지워지지 않으며, 위에 적은 때에 지워지거나 앱을 삭제할 때 함께 삭제됩니다. 리포트 화면의 ‘PDF 저장’도 이 기기 안에서 파일을 만들며, 리포트 내용을 서버로 보내지 않습니다. 내보내거나 PDF로 저장한 파일은 이용자가 직접 관리하여야 합니다.

모바일 앱은 이 저장 공간을 기기 백업(클라우드 백업 포함)에서 제외하도록 설정되어 있습니다. 그래서 기기를 바꾸거나 복원하면 리포트가 따라오지 않으며, 옮기려면 ‘내 데이터 내보내기’를 사용하여야 합니다. 다만 심층 리딩의 구매 정보는 앱의 기본 설정 저장 공간에 있어 이 설정의 대상이 아니므로, 기기 백업에 포함될 수 있습니다.

검토 메모

기기 백업 제외는 두 모바일 앱의 설정 파일과 앱 코드에 들어 있으나 실제 기기에서 확인한 적이 없습니다. iOS에서 웹뷰의 저장 공간이 표시한 폴더에 놓이는지 실제 기기로 확인한 뒤 위 문단을 확정하여야 합니다. 셀프체크의 응답은 건강에 관한 정보일 수 있어 App Store 심사 지침 5.1.3과도 관련됩니다. ‘이 기기의 모든 데이터 삭제’ 뒤에도 게시물 분류값의 임시 보관분이 인스타그램 리포트 화면을 다음에 열 때까지 남을 수 있고, 다시 확인할 수 있게 되는 시각이 지날 때까지 남는 점이 ‘모든 데이터 삭제’라는 표시와 맞는지 검토가 필요합니다. 심층 리딩의 구매 정보(src/lib/billing/iap.ts)는 앱의 기본 설정 저장 공간에 두는데, iOS에서 백업에서 제외한 것은 웹 데이터 폴더(Library/WebKit)뿐이므로 이 정보는 기기 백업에 포함될 수 있습니다. 백업에서 복원된 기기에서 진행 중이던 구매가 다시 쓰일 수 있는지, 이를 백업에서 제외하여야 하는지 확인이 필요합니다. 이 정보가 ‘이 기기의 모든 데이터 삭제’로 지워지지 않는 점이 그 표시와 맞는지도 함께 검토하여야 합니다.

4.만 14세 미만 아동의 개인정보

서비스는 성인만 이용할 수 있습니다. 대한민국에서는 만 19세 이상을 말합니다. 만 14세 미만 아동을 대상으로 하지 않으며, 운영자는 만 14세 미만 아동의 개인정보를 알면서 수집하지 않습니다.

연령은 검사나 인스타그램 리포트를 처음 시작할 때, 그리고 AI 해석에 동의하기 전에 한 번 확인합니다. 이용자가 화면에 생년월일을 입력하면 기기 안에서 만 나이를 계산하며, 입력한 생년월일은 저장하지 않고 운영자에게 전송하지도 않습니다.

성인으로 확인되면 확인을 마쳤다는 사실만 이용자의 기기에 저장됩니다. 기준 연령에 미치지 못하는 생년월일이 입력되면 이용할 수 없으며, 다시 확인할 수 있게 되는 시각(10분 뒤)만 기기에 저장됩니다. 그 시각도 운영자에게 전송되지 않으며, 지난 뒤 앱을 다음에 열 때 지워집니다.

만 14세 미만 아동의 개인정보가 처리된 사실을 알게 된 경우 운영자는 지체 없이 삭제 등 필요한 조치를 합니다.

검토 메모

연령 확인은 이용자가 입력한 생년월일에 기대는 자기 확인 방식이며, 신분증이나 본인확인 수단으로 검증하지 않습니다. 기준에 미치지 못하는 날짜가 입력된 뒤 다시 입력하기까지 10분을 기다리게 하지만, 그 뒤에는 다른 날짜를 넣을 수 있습니다. 설정의 ‘이 기기의 모든 데이터 삭제’로는 이 대기 시간이 풀리지 않으나, 브라우저에서 이 사이트의 데이터를 지우거나 앱을 다시 설치하면 바로 다시 입력할 수 있습니다. 개인정보보호위원회의 아동·청소년 안내서가 든 우회 방지 조치에 비추어 이것으로 충분한지, 만 14세 이상 미성년자의 이용을 이 방식으로 막는 것으로 충분한지 검토가 필요합니다. 연령 확인 전에 만들어진 리포트는 확인 없이 열어 볼 수 있으나, AI 해석을 요청하려면 확인을 거쳐야 합니다.

5.개인정보의 처리 및 보유 기간

정보보유하는 곳보유 기간
AI 해석을 위해 전송된 채점 결과운영자의 서버(메모리)저장하지 않음. 응답을 돌려준 뒤 남지 않음
인스타그램 리포트를 위해 전송된 이미지, 글, 근거 자료운영자의 서버(메모리)저장하지 않음. 응답을 돌려준 뒤 남지 않음
Instagram 로그인 연결로 읽어 온 게시물과 접근 토큰운영자의 서버(메모리)저장하지 않음. 토큰은 이용자의 브라우저에만 있으며 30분 뒤에는 쓸 수 없음
위 내용 가운데 AI 서비스 제공자에게 전달된 내용Anthropic, PBC(미국)받거나 생성한 날부터 30일 이내에 삭제합니다. 다만 이용정책 위반으로 분류된 입력과 출력은 최대 2년, 안전 분류 점수는 최대 7년까지 보관될 수 있습니다.
접속 기록과 AI 요청의 처리 기록(내용 제외)호스팅 사업자(Vercel Inc.)30일
요청 횟수 제한을 위한 해시값과 횟수Upstash, Inc.처음 기록된 때부터 제한 구간의 길이가 지나면 자동 삭제. 요청의 종류에 따라 1분, 10분 또는 하루
심층 리딩 구매를 확인하기 위해 받은 거래 기록이나 구매 토큰운영자의 서버(메모리)저장하지 않음. 확인을 마친 뒤 남지 않음
구매 사용 기록(거래 번호의 해시값, 사용 횟수, 처음 리딩을 받은 시각)Upstash, Inc.400일. 지나면 자동 삭제
문의와 권리 행사 요청의 기록운영자3년
기기에 저장된 응답, 기록, 그림, 리포트이용자의 기기이용자가 삭제할 때까지. 운영자는 보유하지 않음
기기에 임시로 두는 게시물 분류값이용자의 기기7일. 운영자는 보유하지 않음
기기에 두는 심층 리딩의 구매 정보(모바일 앱)이용자의 기기진행 중인 구매는 리포트가 도착할 때까지. 다시 쓰지 않을 구매의 거래 번호는 가장 최근의 것 50개까지. 운영자는 보유하지 않음

심층 리딩의 결제는 앱 마켓이 직접 처리하며, 운영자가 결제에 관하여 보관하는 것은 위 표의 구매 사용 기록뿐입니다.

검토 메모

전자상거래법 제6조와 같은 법 시행령 제6조는 통신판매업자에게 계약 또는 청약철회에 관한 기록과 대금결제 및 재화 등의 공급에 관한 기록을 5년, 소비자의 불만 또는 분쟁처리에 관한 기록을 3년 보존하도록 정하고 있습니다. 결제와 거래 기록을 앱 마켓이 가지고 있고 운영자는 거래 번호의 해시값만 400일 보관하는 구조로 이 의무를 이행할 수 있는지, 환불 요청과 그 처리 기록을 얼마 동안 보관할지(3년) 검토가 필요합니다. 보존 의무가 있다면 그 근거, 항목, 기간을 이 표와 ‘개인정보의 파기 절차 및 방법’에 적습니다.

6.개인정보의 파기 절차 및 방법

운영자는 보유 기간이 지나거나 처리 목적을 달성하여 개인정보가 불필요하게 되었을 때에는 지체 없이 해당 개인정보를 파기합니다.

  • AI 해석과 인스타그램 리포트를 위해 받은 내용(심층 리딩의 구매 확인을 위해 받은 거래 기록과 구매 토큰 포함), Instagram 로그인 연결로 읽어 온 내용: 서버의 메모리에서만 처리하고 디스크나 데이터베이스에 기록하지 않습니다. 결과를 돌려준 뒤에는 서버에 남지 않습니다.
  • 요청 횟수 제한을 위한 값: 만료 시간이 설정되어 있어, 처음 기록된 때부터 제한 구간의 길이가 지나면 자동으로 삭제됩니다.
  • 구매 사용 기록: 만료 시간이 설정되어 있어, 400일이 지나면 자동으로 삭제됩니다.
  • 접속 기록, AI 요청의 처리 기록, 문의 기록: 보유 기간이 지나면 복구하거나 재생할 수 없는 방법으로 삭제합니다.
  • 기기에 저장된 정보: 이용자가 설정에서 직접 삭제합니다.

다른 법령에 따라 보존하여야 하는 개인정보가 있는 경우에는 해당 정보를 다른 개인정보와 분리하여 보관하고, 보존 근거와 항목, 기간을 이 처리방침에 알립니다.

검토 메모

접속 기록에 법정 보존 의무(통신비밀보호법 등)가 적용되는지 확인이 필요합니다. 리서치 문서는 해당 기간(3개월)을 확인되지 않은 기억으로 표시하고 있습니다. 적용된다면 근거와 기간을 이 항목과 보유 기간 표에 적습니다.

7.개인정보의 제3자 제공

운영자는 이용자의 개인정보를 제3자에게 제공하지 않습니다. 개인정보를 판매하지 않으며, 광고 목적으로 이용하거나 제공하지도 않습니다.

아래의 처리업무 위탁과 국외 이전은 운영자의 업무를 대신 처리하도록 맡기는 것으로, 제3자 제공과 구분됩니다.

Instagram 로그인 연결을 이용하는 경우 로그인과 승인은 Instagram(Meta Platforms, Inc.)의 화면에서 이루어지며, 그 과정에서 Instagram이 처리하는 정보에는 Instagram의 개인정보처리방침이 적용됩니다. 운영자는 이 과정에서 이용자의 개인정보를 Instagram에 제공하지 않습니다. Instagram이 발급한 인증 코드와 토큰을 Instagram에 제시하여 이용자 본인 계정의 게시물을 읽어 올 뿐입니다.

앱에서 심층 리딩을 구매하는 경우 결제는 앱 마켓의 화면에서 이루어지며, 그 과정에서 앱 마켓이 처리하는 정보에는 앱 마켓의 개인정보처리방침이 적용됩니다. 운영자는 이 과정에서 이용자의 개인정보를 앱 마켓에 제공하지 않습니다. 앱 마켓이 발급한 구매 토큰을 앱의 식별자, 상품 ID와 함께 그 앱 마켓의 결제 시스템에 제시하여 주문 번호와 구매 상태를 확인하고, 앱 마켓이 요구하는 경우 그 구매가 사용되었음을 알릴 뿐입니다.

검토 메모

구매 토큰을 앱 마켓의 결제 시스템에 보내 구매를 확인하는 것을 개인정보의 제3자 제공이나 국외 이전으로 다룰 필요가 있는지 확인이 필요합니다. 앱 마켓이 발급한 값을 그 앱 마켓에 되돌려 제시하는 것이므로 위와 같이 적었으나, 확인을 받는 결제 시스템은 국외에 있습니다. Instagram 로그인 연결에서 Meta에 인증 코드와 토큰을 제시하는 경우와 같은 기준으로 판단합니다.

8.추가적인 이용·제공의 판단 기준

해당 사항이 없습니다. 운영자는 수집한 목적과 다른 목적으로 개인정보를 추가로 이용하거나 제공하지 않습니다.

9.개인정보 처리업무의 위탁

운영자는 서비스를 제공하기 위하여 다음과 같이 개인정보 처리업무를 위탁하고 있습니다.

수탁자위탁하는 업무처리 위치
Anthropic, PBCAI 해석 문장의 생성과 이를 위한 게시물 이미지·글의 분석(Claude API)미국
Vercel Inc.서비스 서버의 운영(호스팅), 접속 기록과 서버 로그의 보관미국(주된 처리 시설). Vercel과 그 하위 수탁자가 운영하는 다른 국가에서 처리될 수 있음
Upstash, Inc.요청 횟수 제한과 구매 사용 기록을 위한 저장소의 운영미국

운영자는 위탁계약을 맺을 때 「개인정보 보호법」 제26조에 따라 위탁 목적 외 처리의 금지, 기술적·관리적 보호조치, 재위탁의 제한, 수탁자에 대한 관리와 감독, 손해배상 등 책임에 관한 사항을 문서로 정하고, 수탁자가 개인정보를 안전하게 처리하는지 감독합니다. 수탁자에게는 이 처리방침과 관련 법령이 정한 것과 같거나 그보다 높은 수준으로 개인정보를 보호할 것을 요구합니다.

Anthropic, PBC의 재위탁 현황: 클라우드 인프라(Google Cloud Platform, Amazon Web Services, Microsoft Azure, 전 세계)와 트래픽 라우팅(Cloudflare)을 재위탁하며, 데이터는 미국에 저장됩니다. 전체 목록은 https://trust.anthropic.com/subprocessors에서 확인할 수 있습니다.

위탁하는 업무의 내용이나 수탁자가 바뀌는 경우에는 지체 없이 이 처리방침을 통하여 공개합니다.

검토 메모

각 수탁자의 약관과 데이터 처리 부속 계약이 법 제26조제1항과 시행령 제28조제1항의 일곱 가지 항목을 충족하는지 대조하여 기록한 뒤에 위 문단을 확정하여야 합니다. 호스팅 사업자와 저장소 사업자의 재위탁(하위 수탁자) 현황은 아직 적지 않았으므로, 각 사업자가 공개한 목록을 확인하여 추가하여야 합니다. 서버 로그(AI 요청의 처리 기록 포함)를 호스팅 사업자가 아닌 별도의 로그 서비스로 보내게 되면 그 사업자도 이 표와 국외 이전 표에 추가하여야 합니다.

10.개인정보의 국외 이전

운영자는 서비스를 제공하기 위하여 아래와 같이 개인정보 처리업무를 국외에 위탁하고 있으며, 「개인정보 보호법」 제28조의8제2항에 따라 다음 사항을 알립니다.

AI 서비스 제공자: Anthropic, PBC

국외 이전의 법적 근거

「개인정보 보호법」 제28조의8제1항제3호가목: 이용자와의 계약을 이행하는 데 필요한 처리위탁으로서, 같은 조 제2항 각 호의 사항을 이 처리방침에 공개합니다.

성향 리포트의 AI 해석과 인스타그램 리포트 모두, 전송하기 전에 같은 항 제1호에 따른 국외 이전 동의도 함께 받습니다.

이전 유형

개인정보 처리업무의 위탁

이전되는 개인정보 항목

성향 리포트의 AI 해석: 리포트의 종류와 버전, 표시 언어, 척도의 이름과 점수(범위 포함), 백분위·구간, 답한 문항 수, 유형의 이름과 한 줄 설명, 응답 품질에 관한 안내 문장. 문항마다 고른 응답, 검사를 마친 시각, 리포트의 그림을 그리는 데 쓰는 파생 수치는 이전되지 않습니다.

인스타그램 리포트의 분류 단계: 이용자가 고른 본인 계정 게시물의 축소한 이미지와 글(캡션), 게시물 번호, 리포트의 언어. 글 속의 다른 계정 이름(@)은 가린 뒤에 이전합니다.

인스타그램 리포트의 해석 단계: 글자로 된 근거 자료(게시물마다의 번호, 게시 날짜, 종류, 중립적인 한 줄 묘사와 분류값, 글의 앞부분)와 기기에서 계산한 집계값, 그리고 이를 바탕으로 만들어진 해석 문장.

인스타그램 리포트의 첫인상 리딩: 이용자가 고른 본인 계정의 축소한 이미지(11장까지)와 글(캡션), 이미지 번호, 리포트의 언어. 글 속의 다른 계정 이름(@)은 가린 뒤에 이전하며, 게시 날짜와 계정 이름은 이전되지 않습니다.

Human.ai에는 회원 계정이 없으므로 가입 정보(이름, 이메일 주소, 회원 식별자)는 이전되지 않습니다.

이전되는 국가

미국. 운영자는 AI 추론이 미국 안에서만 이루어지도록 요청마다 처리 지역을 미국으로 지정하고, 응답에 표시된 실제 추론 지역을 기록하여 확인합니다.

이전 시기 및 방법

이용자가 AI 해석 또는 인스타그램 리포트를 요청할 때마다, 암호화된 통신(TLS)으로 전송합니다.

이전받는 자(명칭과 연락처)

Anthropic, PBC (연락처: privacy@anthropic.com)

이전받는 자의 이용 목적

AI 해석 문장의 생성과 이를 위한 게시물 이미지·글의 분석(위탁받은 업무의 수행)

이전받는 자의 보유·이용 기간

받거나 생성한 날부터 30일 이내에 삭제합니다. 다만 이용정책 위반으로 분류된 입력과 출력은 최대 2년, 안전 분류 점수는 최대 7년까지 보관될 수 있습니다.

이전을 거부하는 방법, 절차 및 거부의 효과

성향 리포트의 AI 해석: 동의하지 않거나 AI 해석을 요청하지 않으면 이전되지 않습니다. 동의한 뒤에는 설정에서 철회할 수 있습니다. 이 경우에도 기본 해석이 담긴 리포트를 이용할 수 있습니다.

인스타그램 리포트: 국외 이전에 동의하지 않으면 이전되지 않습니다. 동의는 리딩마다 새로 받습니다. 동의하지 않으면 인스타그램 리포트를 이용할 수 없으며, 다른 리포트의 이용에는 제한이 없습니다.

AI 서비스 제공자의 기능 가운데 운영자가 사용하는 것

추론 지역의 지정

모든 요청에 추론 지역을 미국으로 지정하며, 응답에 표시된 실제 추론 지역을 서버의 처리 기록에 남겨 확인합니다.

프롬프트 캐시

한 번의 리딩 안에서 여러 번 읽히는 근거 자료(글자로 된 게시물 분류표)와 모든 이용자에게 같은 지시문을 AI 서비스 제공자의 단기 캐시에 올립니다. 제공자의 문서에 따르면 이 캐시는 기본 5분 동안 유지되고 읽힐 때마다 그만큼 연장되며, 다른 조직과 공유되지 않습니다. 사진과 성향 리포트의 점수는 캐시에 올리도록 지정하지 않습니다.

거절된 요청의 대체 처리

AI 서비스 제공자의 안전 장치가 요청을 거절한 경우, 같은 제공자의 다른 모델이 같은 요청을 한 번 더 처리하도록 하는 기능을 켜 두었습니다. 내용이 다른 사업자에게 넘어가지는 않으며, 리포트에는 해석을 실제로 작성한 모델의 이름이 표시됩니다.

사용하지 않는 기능

파일 보관, 일괄 처리, 피드백 전송처럼 내용을 요청 처리 이후까지 보관하게 하는 기능은 사용하지 않습니다.

호스팅 사업자: Vercel Inc.

국외 이전의 법적 근거

「개인정보 보호법」 제28조의8제1항제3호가목: 이용자와의 계약을 이행하는 데 필요한 처리위탁·보관으로서, 같은 조 제2항 각 호의 사항을 이 처리방침에 공개합니다.

이전 유형

개인정보 처리업무의 위탁, 보관

이전되는 개인정보 항목

접속 기록(IP 주소, 접속 일시, 요청한 주소, 브라우저와 기기 정보)과 AI 요청의 처리 기록(내용 제외)

AI 해석, 인스타그램 리포트(심층 리딩의 구매 정보 포함), Instagram 로그인 연결 요청을 처리하는 동안 서버의 메모리에서 다루는 내용. 저장하지 않습니다.

이전되는 국가

미국(주된 처리 시설). Vercel과 그 하위 수탁자가 운영하는 다른 국가에서 처리될 수 있음

이전 시기 및 방법

서비스에 접속하거나 요청을 보낼 때마다, 암호화된 통신(TLS)으로 전송합니다.

이전받는 자(명칭과 연락처)

Vercel Inc. (연락처: privacy@vercel.com)

이전받는 자의 이용 목적

서비스 서버의 운영, 접속 기록과 서버 로그의 보관

이전받는 자의 보유·이용 기간

30일

이전을 거부하는 방법, 절차 및 거부의 효과

서비스에 접속하지 않는 방법으로 거부할 수 있습니다. 이 경우 서비스를 이용할 수 없습니다.

요청 제한·구매 기록 저장소 사업자: Upstash, Inc.

국외 이전의 법적 근거

「개인정보 보호법」 제28조의8제1항제3호가목: 이용자와의 계약을 이행하는 데 필요한 처리위탁·보관으로서, 같은 조 제2항 각 호의 사항을 이 처리방침에 공개합니다.

이전 유형

개인정보 처리업무의 위탁, 보관

이전되는 개인정보 항목

IP 주소를 비밀값과 함께 일방향 해시 처리한 값과 요청 횟수, 서비스 전체의 하루 AI 요청 수

앱에서 심층 리딩을 구매한 경우: 거래 번호를 일방향 해시 처리한 값과, 그 구매를 쓴 횟수와 처음 리딩을 받은 시각

이전되는 국가

미국

이전 시기 및 방법

AI 해석, 인스타그램 리포트, Instagram 로그인 연결을 요청할 때와 앱이 제공 기능을 확인할 때마다, 암호화된 통신(TLS)으로 전송합니다.

이전받는 자(명칭과 연락처)

Upstash, Inc. (연락처: privacy@upstash.com)

이전받는 자의 이용 목적

요청 횟수 제한을 위한 임시 저장과 구매 사용 기록의 보관

이전받는 자의 보유·이용 기간

요청 횟수 제한 값은 처음 기록된 때부터 제한 구간의 길이가 지나면 자동으로 삭제됩니다. 제한 구간은 요청의 종류에 따라 1분, 10분 또는 하루입니다. 구매 사용 기록은 400일이 지나면 자동으로 삭제됩니다.

이전을 거부하는 방법, 절차 및 거부의 효과

AI 해석, 인스타그램 리포트, Instagram 로그인 연결을 요청하지 않고 심층 리딩을 구매하지 않는 방법으로 거부할 수 있습니다. 이 경우 해당 기능을 이용할 수 없습니다.

운영자는 국외 이전과 관련하여 「개인정보 보호법 시행령」 제29조의10에 따른 보호조치(안전성 확보 조치, 고충 처리와 분쟁 해결에 관한 조치 등)에 관하여 이전받는 자와 미리 협의하고 이를 계약 내용에 반영합니다.

검토 메모

‘미국’ 기재는 모든 AI 요청에 미국 고정이 적용되어 있을 때에만 정확합니다. 코드는 요청마다 지역을 지정하고 응답의 추론 지역을 기록하지만, 실제 API 키로 운영 환경에서 확인한 적은 없고 Anthropic 워크스페이스의 지역 제한은 운영자가 따로 설정하여야 합니다. 거절된 요청을 대체 모델이 처리할 때에도 같은 지역과 보유 조건이 적용되는지, 프롬프트 캐시에 놓인 내용과 대체 처리를 위해 제공자가 남기는 기록이 계약상의 보유 조건에 어떻게 포함되는지는 Anthropic에 확인이 필요합니다. Anthropic이 공개한 하위 수탁자 안내는 클라우드 인프라의 위치를 ‘전 세계’로 적고, 데이터는 미국에 저장한다고 밝히고 있습니다. 추론 지역을 미국으로 지정한 경우에 이전되는 국가를 ‘미국’으로만 적어도 되는지, 하위 수탁자의 국가를 함께 적어야 하는지 판단이 필요합니다. 호스팅 사업자는 주된 처리 시설이 미국에 있으나 자신과 하위 수탁자가 운영하는 다른 국가에서도 처리할 수 있다고 정하고 있으므로, 이 기재가 이전되는 국가를 알리는 요건을 충족하는지도 확인이 필요합니다. 저장소 사업자의 처리 국가는 데이터베이스를 만들 때 정한 주 지역이며, 운영자만 콘솔에서 확인할 수 있습니다. 그 지역이 국내라면 해당 표를 삭제하고 ‘처리업무의 위탁’에만 남기며, 다른 지역에 복제본을 두었다면 그 국가도 함께 적습니다. Instagram 로그인 연결에서 서버가 Meta에 보내는 것은 Meta가 발급한 인증 코드와 토큰뿐인데, 이를 국외 이전으로 다룰 필요가 있는지도 확인이 필요합니다.

검토 메모

서버는 받은 내용을 메모리에서만 처리하고 사본을 두지 않으며 요청 내용을 기록하지 않도록 만들어져 있고, 자동 검사가 그 일부를 지켜봅니다. 다만 이 동작을 운영 환경에서 확인한 적은 없으므로, 첫 배포 뒤 서버 로그를 직접 읽어 이 표와 대조하여야 합니다. 또한 성향 리포트의 AI 해석은 처리 근거를 동의(제15조제1항제1호)로, 국외 이전 근거를 계약 이행을 위한 처리위탁(제28조의8제1항제3호)과 동의(같은 항 제1호)로 함께 적었습니다. 두 근거를 함께 적는 것이 맞는지, 리서치 문서가 권고한 계약 이행(제15조제1항제4호) 구성이 더 적절한지 확인이 필요합니다.

11.생성형 인공지능의 이용에 관한 안내

서비스의 AI 해석은 생성형 인공지능(AI)이 작성합니다. 사람이 직접 작성하거나 전문가가 개별적으로 검토한 결과가 아닙니다. 운영자는 이를 위하여 외부의 생성형 인공지능 서비스(Anthropic, PBC의 Claude API)를 이용합니다. 인스타그램 리포트의 기본 리딩과 심층 리딩은 같은 제공자의 서로 다른 모델을 쓰며, AI 서비스 제공자에게 전달하는 항목은 어느 쪽이나 같습니다. 첫인상 리딩은 심층 리딩과 같은 모델로, 고른 이미지와 글을 한 번의 요청으로 읽습니다.

생성형 AI 이용 요약

의도된 용도

자기이해를 돕는 참고용 해석 문장의 작성. 성향 리포트의 점수에 대한 해석과 인스타그램 리포트에 한정합니다. 진단이나 평가를 위한 것이 아니며, 채용, 진학, 법적 판단 같은 결정의 근거로 쓰기 위한 것도 아닙니다.

AI를 이용하지 않는 리포트

셀프체크는 미리 작성해 둔 문장을 채점 규칙에 따라 보여 주며, AI 서비스로 아무것도 전송하지 않습니다.

입력과 생성 결과의 수집

운영자는 전송받은 내용과 생성된 해석을 서버에 저장하지 않습니다. 생성된 해석은 이용자의 기기에만 저장됩니다.

AI 학습에의 이용

운영자는 이용자의 입력과 생성 결과를 인공지능 모델의 학습이나 개선에 이용하지 않습니다. AI 서비스 제공자의 보유·이용 조건은 개인정보의 국외 이전에 적은 바와 같습니다.

입력과 생성 결과의 보유 기간

운영자: 저장하지 않습니다.

AI 서비스 제공자: 받거나 생성한 날부터 30일 이내에 삭제합니다. 다만 이용정책 위반으로 분류된 입력과 출력은 최대 2년, 안전 분류 점수는 최대 7년까지 보관될 수 있습니다.

학습을 거부하는 방법

운영자가 학습에 이용하지 않으므로 별도의 절차가 없습니다. AI 서비스로 전송되는 것 자체를 원하지 않으면 설정에서 AI 해석 동의를 철회하면 되며, 그 뒤로는 아무것도 전송하지 않습니다. 인스타그램 리포트는 리딩마다 동의를 새로 받으므로 동의하지 않으면 전송하지 않습니다.

생성 결과의 표시

AI가 작성한 해석에는 AI가 생성하였다는 사실을 표시합니다.

생성 결과의 확인

AI가 쓴 문장은 보여 주기 전에 정해진 규칙으로 다시 확인합니다. 점수나 게시물 근거에 연결되지 않은 문장과 리포트가 다루지 않기로 한 주제의 문장은 걸러 내도록 만들어져 있습니다. 규칙이 모든 오류를 잡아내지는 못합니다.

  • AI 해석은 사실과 다르거나 이용자에게 맞지 않을 수 있습니다.
  • 인스타그램 리포트를 이용할 때에는 다른 사람의 개인정보나 민감한 내용이 담긴 게시물을 전송 전에 제외하여 주시기 바랍니다.
  • 부적절하거나 사실과 다른 해석은 contact@craving.win(으)로 알려 주시기 바랍니다.

12.민감정보의 처리와 공개 가능성

  • 셀프체크의 응답과 결과(기분과 주의력에 관한 문항 응답, 주의력 과제의 수행 기록, 그림, 점수)는 건강에 관한 정보에 해당할 수 있습니다. 이 정보는 이용자의 기기 안에서만 계산하고 저장하며, 운영자의 서버로 전송하지 않습니다.
  • 인스타그램 리포트를 위해 이용자가 고른 게시물에는 건강, 사상·신념, 정치적 견해, 성생활 등에 관한 내용이 드러날 수 있습니다. 운영자는 「개인정보 보호법」 제23조제1항제1호에 따라 다른 개인정보의 처리에 대한 동의와 구분하여 별도의 동의를 받은 경우에만 이 내용을 처리합니다.
  • 인스타그램 리포트는 건강이나 정신건강 상태, 성적 지향, 인종·민족, 종교, 정치적 견해, 소득, 지능, 애착 유형, 미성년자에 관한 사항, 사진에 함께 나온 다른 사람에 관한 사항을 추정하지 않도록 설계되어 있습니다.
  • 미성년자, 건강, 성, 추모, 음주·흡연, 자해 표현 등이 감지된 게시물은 어떤 해석의 근거로도 쓰지 않으며, 종교·정치에 관한 게시물은 가치에 관한 해석에서 제외합니다. 그 가운데 미성년자가 보이거나 이름이 나온 게시물, 의료 장면이 담긴 게시물, 자해 표현이 있는 게시물은 올린 날짜를 게시 리듬 계산에 쓰는 것 말고는 리포트의 어떤 수치에도 넣지 않습니다. 다만 그러한 게시물도 감지하려면 분류 단계에서 AI 서비스로 전송됩니다.
  • 분류 단계 없이 바로 시작한 첫인상 리딩(보낼 수 있는 게시물이 12개보다 적을 때)에는 게시물마다 감지하는 단계가 없으므로 위의 게시물별 제외가 적용되지 않습니다. 대신 AI에게 건강, 신체와 외모, 나이, 연애 관계, 신념, 정치적 견해, 출신, 재산, 능력에 관한 내용과, 계정 주인이 아닌 사람이나 미성년자에 관한 내용을 쓰지 말라고 지시하며, 보여 주기 전에 리딩과 같은 문구 확인을 거쳐 통과하지 못한 문장은 지웁니다.
  • 게시물을 분류한 뒤 읽을 수 있는 게시물이 적어 제공하는 첫인상 리딩에서는, 분류 단계에서 어떤 해석의 근거로도 쓰지 않기로 한 게시물(위의 미성년자, 건강, 성, 추모, 음주·흡연, 자해 표현 등이 감지된 게시물)을 보내지 않습니다. 자해 표현이 감지된 경우에는 첫인상 리딩을 제공하지 않고 도움받을 수 있는 연락처를 보여 줍니다.

민감정보의 공개 가능성과 비공개를 선택하는 방법

리포트는 이용자의 기기에만 저장되며, 운영자는 이를 공개하지 않습니다. 이용자가 공유 기능을 직접 사용하는 경우에만 선택한 내용이 다른 사람에게 전달됩니다. 공유하지 않으면 리포트는 이용자만 볼 수 있습니다.

검토 메모

셀프체크는 기기 안에서만 처리됩니다. 이 경우에도 법 제23조의 별도 동의가 필요한지, 운영자가 개인정보처리자로서 ‘처리’하는 것으로 보는지 판단이 필요합니다. 리서치 문서는 서버에 저장하는 구조를 전제로 별도 동의를 권고하였습니다. 인스타그램 리포트의 민감정보 별도 동의는 리딩마다 받고 저장하지 않으므로, 운영자는 그 동의를 받았다는 기록을 보유하지 않습니다.

13.개인정보의 안전성 확보조치

운영자는 개인정보를 안전하게 처리하기 위하여 다음과 같은 조치를 하고 있습니다.

  • 저장하지 않는 구조: 서버는 AI 해석과 인스타그램 리포트를 위해 받은 내용을 메모리에서만 처리하고 저장하지 않습니다. 회원 정보나 리포트를 보관하는 데이터베이스가 없습니다.
  • 전송 구간의 암호화: 기기와 서버 사이, 서버와 AI 서비스 사이의 통신을 암호화(TLS)합니다.
  • 요청 내용의 비기록: 전송된 점수, 이미지, 글, 토큰 등 요청 내용을 서버 로그에 남기지 않습니다. AI 요청에 대해서는 처리 단계, 결과, 모델, 걸린 시간, 토큰 수, 추론 지역, 개수만 기록하며, 숫자와 짧은 식별 낱말이 아닌 값은 기록하기 전에 지웁니다. 오류가 난 경우에는 오류의 종류와 코드 같은 진단 정보만 남기며, 오류 메시지처럼 요청 내용이 섞일 수 있는 값은 남기지 않습니다.
  • 전송 전 최소화: 인스타그램 리포트에 쓰는 이미지는 축소하고 촬영 정보 등 메타데이터를 제거한 것만 전송합니다. 서버는 받은 이미지에 메타데이터가 남아 있으면 다시 인코딩하여 제거한 뒤에 AI 서비스로 넘깁니다. 글 속의 다른 계정 이름(@)은 기기에서 가리고 서버에서 한 번 더 가립니다.
  • 요청 제한에서의 IP 주소 최소화: 요청 횟수 제한에는 IP 주소 원문 대신 일방향 해시 처리한 값을 사용합니다. 접속 기록에는 IP 주소가 남습니다.
  • Instagram 토큰의 비저장: Instagram 로그인 연결에서 받은 접근 토큰을 서버에 저장하지 않습니다. 암호화하여 이용자의 브라우저에만 맡기고, 30분이 지나면 받아들이지 않습니다.
  • 구매 기록의 최소화: 심층 리딩의 구매를 확인한 뒤에는 거래 번호를 일방향 해시 처리한 값과 사용 기록만 보관하며, 앱이 보낸 거래 기록과 구매 토큰은 저장하지 않습니다.
  • 요청의 제한: 요청의 횟수와 크기를 제한하고, 하루에 처리하는 AI 요청의 총량에 상한을 두어 서버를 보호합니다.
  • AI 추론 지역의 지정과 확인: 모든 AI 요청에 추론 지역을 미국으로 지정하고, 응답마다 실제 지역을 기록하여 다른 지역이 나오면 경고를 남깁니다.
  • 기기 백업 제외: 모바일 앱은 리포트가 저장되는 공간을 기기 백업에서 제외하도록 설정되어 있습니다.

검토 메모

관리적 조치(내부 관리계획, 운영·클라우드 콘솔의 접근 통제와 인증, 접속 기록의 점검 등)는 실제로 시행하는 내용만 확인하여 추가합니다. 보유하지 않은 인증은 적지 않습니다.

14.가명정보의 처리

해당 사항이 없습니다. 운영자는 가명정보를 처리하지 않습니다.

15.개인정보 자동 수집 장치의 설치·운영 및 거부

서비스는 아래 두 가지 쿠키만 사용합니다. 어느 쪽에도 개인을 식별하는 정보는 들어 있지 않습니다.

언어 선택 쿠키

이름

humanai-locale

목적

이용자가 고른 언어(한국어 또는 영어)를 기억하여 다음 방문 때 같은 언어로 보여 줍니다.

저장되는 값

언어 코드(ko 또는 en)

보관 기간

1년

설정되는 때

이용자가 언어를 직접 바꿀 때

Instagram 로그인 연결용 쿠키

이름

__Host-humanai-ig-oauth

목적

Instagram 로그인 연결을 시작한 바로 그 브라우저에서만 연결이 완료되도록 확인합니다(보안 목적).

저장되는 값

무작위로 만든 값

보관 기간

최대 10분. Instagram에서 돌아오면 바로 삭제됩니다.

설정되는 때

이용자가 Instagram 로그인 연결을 시작할 때에만

전송 범위

이 쿠키가 있는 동안(최대 10분)에는 서비스로 보내는 요청에 함께 전송됩니다. 암호화된 연결(HTTPS)에서만 저장되고 전송되며, 페이지의 스크립트는 이 쿠키를 읽을 수 없습니다.

쿠키 외에 브라우저의 저장 공간을 다음과 같이 씁니다. 로컬 저장소에는 화면 테마 설정과, 이 기기에 저장된 리포트나 진행 중인 검사가 있는지를 나타내는 표시 하나(처음 화면을 고르는 데 씁니다)를 둡니다. 인스타그램 피드 리딩에서 Instagram 내보내기 페이지를 열면, 다시 왔을 때 바로 파일을 고를 수 있도록 그 시각 하나도 둡니다(계정이나 파일에 관한 정보는 없으며, 파일을 읽거나 7일이 지나면 지웁니다). 응답, 리포트, 설정, 성인 확인과 동의의 기록, 게시물 분류값의 임시 보관분은 IndexedDB에 둡니다. 세션 저장소는 사용하지 않습니다. 모바일 앱에서는 심층 리딩의 구매 정보를 앱의 기본 설정 저장 공간에 둡니다(이용자의 기기에만 저장되는 정보). AI 해석이나 인스타그램 리포트를 요청할 때 보내는 내용(심층 리딩의 구매 증빙 포함)은 처리하는 개인정보의 항목과 법적 근거에 적은 것뿐이며, 그 밖에는 이 가운데 어느 것도 서버로 전송하지 않습니다.

쿠키 저장을 거부하려면 브라우저 설정에서 쿠키를 차단하거나 삭제하면 됩니다. 쿠키를 거부하여도 서비스를 이용할 수 있으며, 언어 선택이 기억되지 않거나 Instagram 로그인 연결을 완료하지 못할 수 있습니다. 설정의 ‘이 기기의 모든 데이터 삭제’를 실행하면 언어 선택 쿠키도 삭제됩니다.

운영자는 광고나 분석을 위한 추적 도구를 사용하지 않습니다. 온라인 행태정보를 수집하지 않으며, 제3자가 이를 수집하도록 허용하지도 않습니다.

운영자는 서비스를 운영하기 위하여 하루 단위의 이용 횟수를 서버에서 셉니다. 다른 곳에서 들어와 웹 페이지를 처음 연 횟수, 인스타그램 리포트 화면을 연 횟수, 리딩과 AI 해석을 마친 횟수, 심층 리딩의 결제와 환불 안내의 건수, 피드 리딩이 서버에서 실패한 횟수, 무료 이용의 하루 상한에 닿아 거절된 횟수이며, 앱(운영체제별)과 웹, 결제 환경(실제 결제와 테스트 결제)으로만 나누어 셉니다. 이 집계는 이용자를 구분하지 않는 숫자로만 남고, IP 주소, 기기나 브라우저의 식별자, 거래 번호, 요청의 내용은 쓰지 않습니다. 집계는 Upstash, Inc.에 두며 40일이 지나면 자동으로 삭제됩니다. 운영자는 이 집계를 하루에 한 번 받아 보고, 리딩이나 AI 해석이 끝나거나 결제가 완료되거나 피드 리딩이 서버에서 실패하거나 환불 안내가 나가거나 하루 상한에 닿았을 때, 그리고 하루 횟수가 10회, 50회, 100회처럼 정해 둔 단계에 이르렀을 때에는 그 사실을 바로 받아 봅니다. 받는 곳은 메신저 서비스 Telegram이며, 알림에는 위의 숫자와 구분만 담깁니다.

16.정보주체와 법정대리인의 권리·의무 및 행사 방법

이용자(정보주체)는 운영자에 대하여 언제든지 개인정보의 열람, 정정·삭제, 처리정지를 요구하고 동의를 철회할 수 있습니다.

기기에 저장된 정보

응답과 리포트는 이용자의 기기에만 있으므로 이용자가 직접 확인하고 지울 수 있습니다.

  • 열람: 보관함에서 리포트를 확인하거나, 설정의 ‘내 데이터 내보내기’로 사본을 받습니다.
  • 삭제: 설정의 ‘이 기기의 모든 데이터 삭제’를 실행합니다.
  • 동의 철회: 설정에서 AI 해석 동의를 철회합니다. 즉시 적용되며, 그 뒤로는 아무것도 전송하지 않습니다. 인스타그램 리포트의 동의는 저장되지 않으며, 진행 중인 리딩은 그 화면에서 취소할 수 있습니다.

운영자가 처리하는 정보

접속 기록 등 운영자가 처리하는 정보에 대한 요구는 서면, 이메일 또는 전화로 할 수 있습니다. 접수처: 이메일 contact@craving.win, 전화 010-9027-0729

  • 열람 요구에는 10일 이내에 응답합니다.
  • 정정·삭제 요구에는 지체 없이 필요한 조치를 하고 그 결과를 알립니다.
  • 처리정지 요구에는 지체 없이 처리를 정지하고, 10일 이내에 그 결과를 알립니다.
  • 정보주체가 아닌 사람으로부터 수집한 개인정보에 대하여 고지를 요구받으면 3일 이내에 수집 출처, 처리 목적, 처리정지를 요구하거나 동의를 철회할 권리가 있다는 사실을 알립니다.

운영자는 계정을 두지 않으므로, 요구를 처리하려면 해당 기록이 본인의 것임을 확인할 수 있는 정보(예: 접속 일시와 당시의 IP 주소, 구매 사용 기록이라면 구매한 앱 마켓과 주문 번호)가 필요할 수 있습니다. 확인에는 필요한 최소한의 정보만 요청합니다.

권리 행사는 법정대리인이나 위임을 받은 대리인을 통해서도 할 수 있습니다. 이 경우 대리 권한을 확인할 수 있는 서류를 제출하여야 합니다.

운영자는 권리를 행사하는 방법과 절차를 개인정보를 수집하는 방법과 절차보다 어렵지 않게 운영합니다.

본인의 게시물이 다른 사람에 의해 제출되었다고 생각되는 경우의 요청 방법은 분석 제외 및 개인정보 요청 안내에서 확인할 수 있습니다.

검토 메모

AI 서비스 제공자가 보유 기간 동안 가지고 있는 입력과 생성 결과에 대하여 열람이나 삭제 요구를 받았을 때, 운영자가 해당 요청을 특정하여 수탁자에게 조치를 요구할 수 있는지 계약과 실제 절차로 확인하여야 합니다. 계정이 없고 요청별 식별값을 남기지 않는 구조이므로, 가능한 범위와 방법이 정해지면 이 항목에 적습니다.

17.자동화된 결정

해당 사항이 없습니다. 리포트는 이용자에게 제공하는 참고 정보이며, 운영자는 완전히 자동화된 시스템으로 개인정보를 처리하여 이용자의 권리나 의무에 중대한 영향을 미치는 결정을 하지 않습니다.

18.개인정보 보호책임자

운영자는 개인정보 처리에 관한 업무를 총괄하여 책임지고, 개인정보 처리와 관련한 이용자의 불만 처리와 피해 구제를 위하여 아래와 같이 개인정보 보호책임자를 지정하고 있습니다.

개인정보 보호책임자

성명 또는 직책

송현섭 (대표자)

연락처

contact@craving.win, 010-9027-0729

개인정보 문의 접수

contact@craving.win

운영자

상호

크레이빙(CRAVING)

대표자

송현섭

주소

서울특별시 성북구 안암로 145, 경영본관동 2층 224호(안암동5가, 고려대학교안암캠퍼스)

전화

010-9027-0729

이용자는 서비스를 이용하면서 생긴 개인정보 보호 관련 문의, 불만 처리, 피해 구제에 관한 사항을 개인정보 보호책임자에게 문의할 수 있으며, 운영자는 지체 없이 답변하고 처리합니다.

19.권익침해 구제방법

이용자는 개인정보 침해에 대한 구제를 받기 위하여 아래 기관에 분쟁 해결이나 상담 등을 신청할 수 있습니다.

기관전화홈페이지
개인정보분쟁조정위원회(국번 없이) 1833-6972www.kopico.go.kr
개인정보침해신고센터(국번 없이) 118privacy.kisa.or.kr
경찰청 사이버범죄신고시스템(국번 없이) 182ecrm.police.go.kr

20.개인정보 처리방침의 변경

이 개인정보 처리방침은 2026년 11월 1일부터 적용합니다.

내용을 추가, 삭제, 수정하는 경우에는 변경 사항을 시행하기 전에 서비스 화면을 통하여 알리고, 변경 전후의 내용을 비교하여 확인할 수 있도록 이전의 처리방침을 함께 공개합니다.

이전의 개인정보 처리방침: 없음(최초 시행)

Policies

Privacy Policy

Service
Human.ai (휴먼ai)
Effective date
1 November 2026

크레이빙(CRAVING) (the “operator”) runs Human.ai (휴먼ai, the “service”). Under Article 30 of Korea's Personal Information Protection Act, the operator publishes this privacy policy to protect your personal information and to handle any related complaint promptly.

In this policy, the “style reports” are Romantic style report, Relationship security report and Interpersonal style report; the “self-checks” are Mood check-in and Focus style report; and the “Instagram report” is Instagram feed reading. The mobile apps downloaded from the app stores do not contain the self-checks, and what this policy says about them applies only where they are offered.

Draft pending legal review

This document is a draft that a lawyer has not yet reviewed, and it is not final. Bracketed text with a dashed outline marks information the operator still has to supply, and each “Review note” is guidance for the reviewer.

  • Human.ai has no accounts and no sign-in. The operator runs no database of your answers or reports.
  • The service is for adults only (19 or over in Korea). You are asked for your date of birth the first time you start, but it is used only to work out your age and is neither stored nor sent; only the fact that you confirmed is kept, on your device.
  • Your answers, task records, drawings and finished reports are stored only on your device.
  • The self-checks are calculated entirely on your device. Nothing about them is sent to a server.
  • An AI interpretation is written only when you consent and ask for one. Only what it needs is then sent through the operator's server to an AI service in the United States, and the operator's server keeps no copy. How long the AI service provider holds what it receives is stated under Overseas transfer of personal information.
  • If you buy the Instagram report's deep reading in the app, the app store handles the payment and the operator receives no name, email address or payment details. Once the purchase has been confirmed as real, only a hash of the transaction number and a record of its use are kept, for 400 days.
  • No advertising or analytics trackers are used. There are two cookies only: one remembers your language, and one security cookie exists just while you are connecting with Instagram Login.

1.Why personal information is processed

The operator processes the minimum personal information needed for the purposes below. It is not used for anything else, and if a purpose changes the operator will take the steps the law requires, such as asking for separate consent.

  • Providing AI interpretation: writing the AI interpretation you request for the style reports.
  • Providing the Instagram report: reading the posts you chose from your own account and writing a report on the picture your feed presents. It comes as a free basic reading and first look, and as a paid deep reading bought in the app.
  • Confirming purchases: checking that a deep reading bought in the app was really paid for, and making sure one purchase is not used for many readings.
  • Keeping the service stable and secure: managing access logs and processing records of AI requests (without their content), limiting excessive requests and daily usage, and responding to faults and misuse.
  • Handling enquiries and rights requests: checking and answering enquiries sent by email or phone (refund requests included) and requests for access, correction, deletion or suspension of processing.

2.What is processed, and on what legal basis

Processed with your consent

AI interpretation of the style reports

Items

What the operator's server receives: your scored results (scale scores with their ranges, percentiles and bands, how many questions you answered, your type, response-quality indicators, and the time you finished the test), derived numbers the report uses to redraw its charts and tables (depending on the report: coordinates and detailed scores, the codes of neighbouring types, which form of the test and which comparison group were used, how many questions were answered, which scales could not be scored and the numbers of any questions left unanswered, the longest run of identical answers and how often the most-chosen answer was picked, how many pairs of opposite questions were answered inconsistently, and the median time per question), plus the kind and version of the report and the display language. Your answers to individual questions are not sent.

Of that, what is passed on to the AI service provider: the kind and version of the report, the display language, each scale's name and score (with its range), percentiles and bands, how many questions you answered, the name and one-line description of your type, and the notes on response quality. The time you finished the test and the derived numbers above are not used by the server and are not passed on. The wording of the questionnaire itself, which is the same for everyone, is given alongside as the explanation of what each score means.

Purpose

Writing and providing the AI interpretation

Legal basis

Your consent (Personal Information Protection Act, Article 15(1)1)

When

When you have agreed to AI interpretation and ask for one in a report

Record and withdrawal of consent

The fact that you consented, and when, is recorded on your device only and is not sent to the operator. Consent is taken only on a device that has passed the adult check. You can withdraw at any time in Settings, and nothing is sent after that.

Retention

Not stored on the operator's server. It is handled in memory while the request is processed and is not kept once the result has been returned to your device.

Retention and use by the AI service provider acting as the operator's processor: Deleted within 30 days of receipt or generation. Inputs and outputs flagged as a Usage Policy violation may be kept for up to 2 years, and safety classification scores for up to 7 years.

Your right to refuse

You may refuse. You can still use the report with the built-in interpretation that follows from your scale scores.

Instagram report

Items

Coding the posts: downscaled images (768 pixels at most on the long edge, with capture details and other metadata removed) and captions of the posts you chose from your own account, the number each post is given in order, the report language, the kind of material (the post's own picture or a screenshot) and which of the three ways it came by (screenshots, an export file or the Instagram Login connection). For a post of several photos handed over with an export file, up to four photos after the cover (512 pixels at most on the long edge, metadata removed) are sent with it, each labelled with the post's number and its place in the post. Other accounts' names (@) in your captions are masked before sending. The server uses the way the material came by only to mask those names and does not pass it on to the AI service provider.

Reading them: the pictures are not sent again, only evidence in text form. For each post, its number, its date (when you handed posts over with an export file or the Instagram Login connection), the kind of post, the neutral one-line description and the categories the AI assigned when coding (including any words it transcribed from the picture), and the start of the caption (up to 300 characters); and counts worked out on your device (numbers and shares of posts, the six axis scores and the feed type, and posting-rhythm and colour statistics).

The first look (a free reading offered when you have fewer than 12 posts that can be sent, or when fewer than that could be read after coding): it is sent in one request, without separate coding and reading steps. Up to 11 images you chose from your own account are sent, downscaled and stripped of metadata in the same way as for coding, with the number each image is given, its caption if it has one, the report language and the way the material came by. Other accounts' names (@) in captions are masked before sending, and the server uses the way the material came by only to mask those names. No dates and no counts are sent. When it is offered after coding, posts the coding step kept from every interpretation are not sent.

If you choose screenshots, whatever is visible in them (an account name, profile bio, follower and like counts, other people's comments and so on) is sent as part of the picture. Apart from that, your account name, profile bio, and like, comment and follower counts are not sent to the server. If you connect with Instagram Login, however, the server itself reads some of these from your account (see the table below), and even then your account name, bio, and like and comment counts are not passed on to the AI service provider. Posts you leave out are not sent, picture or caption; only their dates count toward the posting-rhythm figures.

How you hand it over

Screenshots or photos you pick from your own account (feed posts, a single slide of a carousel, stories and highlights, captures of your profile screen), your own “Export your information” file downloaded from Instagram and read on your device, or, where offered, the Instagram Login connection described below.

The export file is read on your device. The file itself is not sent to the server.

Sensitive information

Posts can contain things that reveal sensitive information, such as health or beliefs. For that reason a consent separate from every other consent is requested.

Purpose

Writing and providing the Instagram report

Legal basis

Your consent (Personal Information Protection Act, Article 15(1)1)

Your separate consent to the processing of sensitive information (Article 23(1)1)

Your separate consent to the overseas transfer (Article 28-8(1)1)

How consent is taken

Each time you start a reading you are asked for four things, each on its own: confirmation that the material is from your own account, consent to collection and use, consent to the processing of sensitive information, and consent to the overseas transfer. These consents apply to that reading only and are not stored anywhere. A first look asks for the same four. The consents are asked on the last screen, which shows the photos and captions to be sent (or, when you connect with Instagram Login, before you connect), and even after you agree, nothing is sent until you press start on that screen, and if you cancel after sending has started, nothing more is sent. The count-only report from an export file, which is offered when the AI service is unavailable, sends nothing, so only the confirmation that the material is from your own account is asked.

Retention

Not stored on the operator's server. It is processed in memory only, and the result is returned to your device.

Retention and use by the AI service provider acting as the operator's processor: Deleted within 30 days of receipt or generation. Inputs and outputs flagged as a Usage Policy violation may be kept for up to 2 years, and safety classification scores for up to 7 years.

So that the same post is not sent twice, the categories and one-line description the AI assigned to each post (including a caption read from a screenshot) are kept on your device for 7 days. No picture is stored there.

Your right to refuse

You may refuse. The Instagram report is then unavailable; every other report still works.

Instagram Login connection (optional, on the web only)

When this applies

Only when you have a professional account and connect it yourself. While the operator does not offer this function, it does not apply.

Permission requested

A single basic read permission (instagram_business_basic). No permission to publish, message, comment or read insights is requested.

Items

The single-use authorisation code and the short-lived access token Instagram issues, your username and display name, your profile bio, and for your recent posts (within the last 36 months, 60 at most) the caption, date, kind of post and number of photos in an album, cover image, and like and comment counts

How it is processed

The operator's server uses your token to read the above from Instagram, downscales the cover images and passes everything to your device. Nothing is looked up by username: only the account the token belongs to, which is yours, is read.

The server does not store the token, the posts or the images. The token is encrypted and left with your browser only, and the server stops accepting it after 30 minutes. Your browser holds it only in the memory of the open page, not in storage or a cookie, so reloading or closing the page loses it.

The address Instagram sends you back to carries the single-use authorisation code, and like any other request that address can appear in the hosting provider's access log. The code can be used only once, and the server exchanges it for the token as soon as it arrives.

The operator never receives your Instagram password. You sign in and approve on Instagram's own screen.

Purpose

Bringing the posts of your own account to your device for the Instagram report

Legal basis

Your consent (Personal Information Protection Act, Article 15(1)1). It is processed only when you start the connection yourself and approve it on Instagram's consent screen.

Retention

Not stored on the operator's server.

Your right to refuse

You do not have to connect. You can still use the Instagram report with screenshots or photos, or with your “Export your information” file.

Disconnecting

When you disconnect in the service, your browser discards the connection it was holding at once. It is also discarded automatically when a reading finishes or you leave that screen. There is no token on the server to erase. The token Instagram issues lasts one hour, and the operator never exchanges it for a longer-lived one or renews it.

How to ask for deletion

Nothing read from Instagram remains on the operator's server. You can delete the reports stored on your device yourself in Settings, and send any other request to contact@craving.win.

Review note

The items sent to the AI service were written from the request format (src/lib/contracts.ts) and the server code (src/server/ai), and automated checks (legal.test.ts) watch some of them. What remains for review is this. A screenshot can show other people's account names and comments, and these are sent as part of the picture, so a judgment is needed on whether that is processing of the personal information of people who are not the data subject. The first look also accepts captures of stories, highlights and the profile screen, so this applies to wider material. The AI service provider's retention (Deleted within 30 days of receipt or generation. Inputs and outputs flagged as a Usage Policy violation may be kept for up to 2 years, and safety classification scores for up to 7 years.) and contact were copied from Anthropic's published help pages (checked 20 October 2026), and the consent screens print the same figures. Confirm them against the commercial terms and data processing addendum the operator accepted; if they differ, the two must be made to say the same thing. The consent screens also say that anything the law requires to be kept may be kept for as long as it requires, and that nothing is used to train AI models, so confirm those two points against the contract as well and decide whether this policy should state them too. The consent screen for the style reports lists the time the test was finished among the items transferred, but the server does not pass it on to the AI service provider. The record of consent exists only on the user's device (the consents for the Instagram report are kept nowhere) and the operator holds no proof that consent was given, so whether this design can meet the burden of proving consent needs review. The scored result of Romantic style report can include scores about the user's current romantic relationship (intimacy, passion, commitment); take that into account when confirming that the style reports are not sensitive information.

Review note

The Instagram Login connection is subject to Meta's app review and Platform Terms, so check that this policy covers what section 4 of the Platform Terms requires (the data processed, how and why, and how to request deletion). The connection screen and the server code exist, but they have never been run against a real Meta app. Whether captions are actually provided over this connection is not confirmed either, so before the operator switches the function on, the items above must be checked against a real response.

Processed without consent

The following is needed to provide the service and run it safely, so it is processed without consent.

Access logs and the server's processing records

Items

IP address, time of access, the address requested, browser and device details: the ordinary access log a hosting provider's servers keep.

Processing records of AI requests: the name of the processing step, the outcome (succeeded, declined, cut off, malformed, failed) and the category of any error, the name of the model used and why its response ended, the time taken, input, output and cache token counts, the region where inference ran, and how many posts, images, statements, questions and report chapters were handled. The record can hold only numbers, true or false, and short identifier-like words, so the content of requests and responses, such as scores, pictures, captions and generated text, does not appear in it.

When the server hits an unexpected error, only the kind of error (its name) is recorded. The error message and where it arose are not recorded, because they can quote the request. When the Instagram Login connection hits an error, the category of the error and the status code and error code Instagram returned are recorded.

Purpose

Providing the service, responding to faults, security

Legal basis

Personal Information Protection Act, Article 15(1)4 (providing the service you asked for) and Article 15(1)6 (legitimate interest in running the service safely)

Retention

30 days

Rate-limit counter

Items

A one-way hash of your IP address, made with a secret value, and a count of requests. The IP address itself is not stored for this purpose. Counters of the whole service's AI requests per day, which distinguish no one, are kept as well.

Purpose

Limiting how many AI interpretation, Instagram report and Instagram Login requests can arrive in a short time, and capping the total number of AI requests handled in a day

Legal basis

Personal Information Protection Act, Article 15(1)4 (providing the service you asked for) and Article 15(1)6 (legitimate interest in running the service safely)

Retention

Each counter carries an expiry time and is deleted automatically once the length of its limit window has passed since it was first written. The windows are 10 minutes and one day for AI interpretation and Instagram report requests (the Instagram report's daily limit applies to the basic reading and the first look only), 10 minutes for Instagram Login requests, and 1 minute for the request that checks which features are available.

Confirming a deep reading purchase

When this applies

Only when you have bought the Instagram report's deep reading in the app. It does not apply to the free basic reading.

Items

What the app sends the server: only what the app store you bought from gave the app for that purchase. Depending on the store, it is one of two things. One is the transaction record the store has signed, sent whole, as the store issued it. Besides the transaction number, the app and product, the time of purchase and whether it was a test or a real payment, it carries the original transaction number and the app transaction number, the store country, the price and currency, values that verify the device, and whatever else the store puts in the record. The other is the product ID and the purchase token. It is sent with each coding request and with the reading request of a deep reading.

A signed transaction record is checked by the server itself, in memory, by verifying its signature, without contacting the app store. Of its contents, the check uses only the transaction number, the app and product, the kind of product, the payment environment, the times of purchase and signing, and whether it was refunded; the rest is not used. Where it receives a purchase token instead, the server sends it, with the app's identifier and the product ID, to that app store's payment system to confirm the order number and the purchase state, and after the reading has been delivered it reports that the purchase has been used.

What the server keeps: a one-way hash of the app store's identity together with the transaction number, and how many readings, reading attempts and coding requests the purchase has had, with when it first produced a reading.

Not received

Your name, email address, payment details such as a card number, and app store account details are not received. The app store handles the payment itself.

Purpose

Confirming that the purchase was really made, and making sure one purchase is not used for many readings

Legal basis

Personal Information Protection Act, Article 15(1)4 (performing the contract with you) and Article 15(1)6 (legitimate interest in keeping one purchase from being used many times)

Retention

The hash of the transaction number and the record of its use: kept for 400 days in the store used for rate limiting (Upstash, Inc.), then deleted automatically. While a reading runs, a lock marker is kept for at most 5 minutes so that two readings cannot run on the same purchase at once.

The transaction record and purchase token the app sends: not stored. They are handled in server memory only while they are checked.

Enquiries and rights requests

Items

The sender's email address if you write by email, the caller's phone number if you call (010-9027-0729), and the content of the enquiry

For an enquiry about a refund or a purchase, the purchase details you give (the app store you bought from and the order number). An order number can be matched to the purchase use record described under “Confirming a deep reading purchase”, so giving it lets the operator find that purchase's use record.

Purpose

Replying to enquiries and refund requests; checking and handling rights requests

Legal basis

Personal Information Protection Act, Article 15(1)4 (taking steps at the data subject's request)

Retention

3 years

Review note

The legal bases given for access logs, processing records of AI requests, the rate-limit counter and enquiry records are the drafter's proposal. The research brief does not address them directly, so they need review before they are final. How long the hosting provider (Vercel Inc.) keeps runtime logs (the processing records of AI requests included) depends on the plan: 1 hour on Hobby, 1 day on Pro, and 30 days on Pro with Observability Plus. The operator's team is on Pro with Observability Plus on (checked 27 October 2026), so the access-log retention is given as 30 days; if Observability Plus is turned off or this project is excluded from it, that figure must become 1 day. If phone enquiries are recorded or written down, add those items and how long they are kept to the enquiry records. If the secret mixed into the hash (APP_SECRET) is not set in production, the IP address can easily be recovered from the hash, so confirm that it is set. An unexpected error is recorded by its name only, so confirm from the production logs that nothing else about it is written. The legal bases given for confirming purchases (performing the contract and legitimate interest), and the statement that the transaction record and purchase token are not stored, were written from the payment code (src/server/billing) but have not been confirmed in production with a real purchase, so read the server logs and the store directly after the first sale.

What is not collected

  • Human.ai has no sign-up or sign-in, so no registration details such as a name, user ID, password or phone number are collected.
  • The app store handles payment itself, so your name, email address, payment details such as a card number, and app store account details are not received. What is received to confirm a purchase is only what is listed under “Confirming a deep reading purchase” above.
  • Your answers to individual questions in the style reports are not sent to the server.
  • The self-checks are calculated on your device only. None of the answers, task records, drawings, scores or results are sent to the server.
  • The date of birth you enter for the adult check is used on your device only to work out your age. It is neither stored nor sent to the server.
  • Nobody's Instagram account is looked up by username. Posts are not collected automatically (scraped), and scraped data is not bought.
  • When your “Export your information” file is read, the folders holding direct messages, contacts, login and device records, and search and advertising history are never opened, and location details attached to posts are not read. Other accounts' names (@) inside your text are replaced with numbers before it is read.
  • No location is requested. Help contacts are chosen for where you are by reading the time zone set on your device, on the device, and that time zone is not sent to the server.

3.What is stored only on your device

The following is stored only on your device. The operator's server holds no copy, and the operator cannot access it.

  • Your answers and the state of any test in progress
  • Records of the attention tasks (response times and accuracy, plus device details such as how you responded and the screen size), and drawings
  • Finished reports. A report includes the AI interpretation if you received one, and an Instagram report includes small preview images and caption excerpts of the posts or images it cites as evidence, plus your account name when it was read from an export file or the Instagram Login connection.
  • The fact that you passed the adult check, the time from which the check can be made again if a date of birth under the required age was entered, whether you allowed AI interpretation and when, your appearance (theme) and haptics settings, and whether you have already seen the introduction shown when the app is first opened. The date of birth itself is not stored.
  • The categories and one-line description the AI assigned to each post when an Instagram report was made. They are kept for 7 days so that the same post is not sent twice, and no picture is stored. If you stop a reading part-way or leave the reading screen, the categories received in that reading are removed with it.
  • Purchase details, if you bought a deep reading in the mobile app. The proof of the purchase a reading is running under (the transaction record the store signed, or the purchase token) and its transaction number are kept until the report arrives, so that the reading can carry on with that purchase after the app is closed and reopened and you can ask again if the result did not reach you. They are removed once the report is delivered or the purchase is closed or set aside. The transaction numbers of purchases the server would not accept, or that used up their attempts without a report being delivered, are kept, the 50 most recent at most, so that those purchases are not used again. Neither holds your name, email address or payment details.

It is kept in your device's browser or app storage (IndexedDB). The theme, a single marker of whether this device has any saved report or test in progress, and the time you went to request your Instagram export are kept in the browser's local storage. The deep reading purchase details are kept in the app's own preferences storage, inside the mobile app only, and never on the web. None of it appears on another device or browser, and it is deleted if you uninstall the app or clear this site's data in your browser.

In Settings you can take a copy of your reports, tests in progress and settings with “Export my data”, and remove your reports, tests in progress, adult confirmation and consent, settings, theme and language choice in one step with “Delete everything on this device”. The temporary store of post categories, the marker of whether this device has any saved report or test in progress, and the time you requested your Instagram export are cleared on the spot as well. Two things are handled differently. If a date of birth under the required age was entered and the time from which the check can be made again has not yet passed, that time stays until it passes. The deep reading purchase details are not removed by this function; they are removed at the times given above, or when you uninstall the app. “Save PDF” on a report also makes the file on this device and sends nothing of the report to the server. Files you export or save as PDF are yours to look after.

The mobile apps are set to keep this storage out of device backups (cloud backups included). Your reports therefore do not follow you when you change or restore a device; to move them, use “Export my data”. The deep reading purchase details, however, sit in the app's preferences storage, which this setting does not cover, so they can be included in a device backup.

Review note

The device backup exclusion is in the configuration files and app code of both mobile apps, but it has never been checked on a real device. Confirm on a real device that the iOS web view keeps its storage in the folder that is marked, then finalise the paragraph above. Self-check answers may be health information, so App Store Review Guideline 5.1.3 is relevant too. Whether the label “Delete everything” is acceptable when the temporary store of post categories can remain until the Instagram report screen is next opened, and a running retry time remains until it passes, needs review. The deep reading purchase details (src/lib/billing/iap.ts) are kept in the app's preferences storage, and on iOS only the web data folder (Library/WebKit) is excluded from backup, so these details can be included in a device backup. Confirm whether a purchase that was in progress could be used again on a device restored from a backup, and whether the details should be excluded from backup. Whether it fits the label “Delete everything on this device” that this function leaves them in place needs review as well.

4.Children under 14

The service is for adults only. In Korea that means 19 or over. It is not directed at children under 14, and the operator does not knowingly collect the personal information of children under 14.

Your age is checked once: the first time you start a test or the Instagram report, and before you consent to AI interpretation. You enter your date of birth on screen and your age is worked out on your device. The date you enter is not stored and is not sent to the operator.

If you are an adult, only the fact that you passed the check is stored, on your device. If the date of birth entered is under the required age, the service cannot be used, and only the time from which the check can be made again (10 minutes later) is stored on the device. That time is not sent to the operator either, and it is removed the next time the app is opened after it has passed.

If the operator learns that personal information of a child under 14 has been processed, it will delete it or take whatever other step is needed without delay.

Review note

The age check relies on the date of birth the user enters. It is a self-declaration and is not verified against an identity document or an identity service. After a date under the required age, the user must wait 10 minutes before entering one again, but can then enter a different date. “Delete everything on this device” in Settings does not lift the wait, but clearing this site's data in the browser or reinstalling the app does, at once. Whether this is enough when set against the anti-circumvention measures in the regulator's guide on children and young people, and whether it is enough to keep out minors aged 14 and over, needs review. A report made before the age check existed can be opened without it, but asking for an AI interpretation requires passing it.

5.How long personal information is kept

InformationHeld byKept for
Scored results sent for AI interpretationThe operator's server (memory)Not stored. Gone once the response is returned
Images, captions and text evidence sent for the Instagram reportThe operator's server (memory)Not stored. Gone once the response is returned
Posts read through the Instagram Login connection, and the access tokenThe operator's server (memory)Not stored. The token exists only in your browser and cannot be used after 30 minutes
The part of the above passed on to the AI service providerAnthropic, PBC (United States)Deleted within 30 days of receipt or generation. Inputs and outputs flagged as a Usage Policy violation may be kept for up to 2 years, and safety classification scores for up to 7 years.
Access logs and processing records of AI requests (without content)Hosting provider (Vercel Inc.)30 days
Hashed value and request count for rate limitingUpstash, Inc.Deleted automatically once the length of the limit window has passed since it was first written: 1 minute, 10 minutes or one day, depending on the request
Transaction record or purchase token received to confirm a deep reading purchaseThe operator's server (memory)Not stored. Gone once the check is done
Purchase use record (hash of the transaction number, times used, time of the first reading)Upstash, Inc.400 days, then deleted automatically
Records of enquiries and rights requestsThe operator3 years
Answers, records, drawings and reports on your deviceYour deviceUntil you delete them. The operator does not hold them
Post categories kept temporarily on your deviceYour device7 days. The operator does not hold them
Deep reading purchase details on your device (mobile app)Your deviceThe purchase in progress: until the report arrives. Transaction numbers of purchases not to be used again: the 50 most recent at most. The operator does not hold them

The payment for a deep reading is handled by the app store itself; the only thing the operator keeps about it is the purchase use record in the table above.

Review note

Article 6 of the E-Commerce Act and Article 6 of its Enforcement Decree require a mail-order business to keep records of contracts and withdrawals, and of payment and supply, for 5 years, and records of consumer complaints and dispute handling for 3 years. Whether that duty can be met when the app store holds the payment and transaction records and the operator keeps only a hash of the transaction number for 400 days, and how long refund requests and their handling are to be kept (3 years), needs review. If a retention duty applies, state its basis, items and period in this table and under “How personal information is destroyed”.

6.How personal information is destroyed

When a retention period ends or the purpose has been achieved and the personal information is no longer needed, the operator destroys it without delay.

  • Content received for AI interpretation and the Instagram report (including the transaction record and purchase token received to confirm a deep reading purchase), and content read through the Instagram Login connection: processed in server memory only and never written to disk or a database. Nothing remains on the server after the result is returned.
  • Rate-limit counter: carries an expiry time and is deleted automatically once the length of its limit window has passed since it was first written.
  • Purchase use record: carries an expiry time and is deleted automatically after 400 days.
  • Access logs, processing records of AI requests and enquiry records: deleted, in a way that cannot be recovered or restored, when the retention period ends.
  • Information on your device: you delete it yourself in Settings.

Where another law requires personal information to be kept, the operator stores it separately from other personal information and states the legal basis, the items and the period in this policy.

Review note

Check whether a statutory retention duty (for example under the Protection of Communications Secrets Act) applies to access logs. The research brief marks the three-month period as recalled, not verified. If a duty applies, state its basis and period here and in the retention table.

7.Provision to third parties

The operator does not provide your personal information to third parties. It does not sell personal information, and does not use or provide it for advertising.

The entrusted processing and overseas transfer described below are arrangements under which a processor works on the operator's behalf. They are distinct from provision to a third party.

If you use the Instagram Login connection, you sign in and approve on the screen of Instagram (Meta Platforms, Inc.), and what Instagram processes in the course of that is governed by Instagram's own privacy policy. The operator does not provide your personal information to Instagram in the process. It only presents to Instagram the authorisation code and token that Instagram itself issued, in order to read the posts of your own account.

If you buy a deep reading in the app, the payment takes place on the app store's own screen, and what the app store processes in the course of that is governed by the app store's own privacy policy. The operator does not provide your personal information to the app store in the process. It only presents the purchase token the app store itself issued, with the app's identifier and the product ID, to that store's payment system, to confirm the order number and the purchase state and, where the store requires it, to report that the purchase has been used.

Review note

Confirm whether sending a purchase token to the app store's payment system to check a purchase needs to be treated as provision to a third party or as an overseas transfer. It is written as above because the value the store itself issued is presented back to that store, but the payment system that answers is outside Korea. Apply the same test as for presenting the authorisation code and token to Meta in the Instagram Login connection.

8.Additional use or provision

Not applicable. The operator does not use or provide personal information for any purpose other than the one it was collected for.

9.Processors (entrusted processing)

To provide the service, the operator entrusts the processing of personal information as follows.

ProcessorEntrusted workWhere it is processed
Anthropic, PBCGenerating AI interpretation text, and analysing post images and captions for it (Claude API)United States
Vercel Inc.Running the service's servers (hosting) and keeping access logs and server logsUnited States (primary processing facilities). May be processed in other countries where Vercel and its sub-processors operate
Upstash, Inc.Running the store used for rate limiting and for purchase use recordsUnited States

When it concludes a processing agreement, the operator sets out in writing, as Article 26 of the Personal Information Protection Act requires, the ban on processing beyond the entrusted purpose, technical and managerial safeguards, limits on sub-processing, supervision of the processor, and liability including damages, and it supervises whether the processor handles personal information safely. Processors are required to protect personal information to a standard equal to or higher than this policy and the applicable law.

Sub-processing by Anthropic, PBC: Cloud infrastructure (Google Cloud Platform, Amazon Web Services, Microsoft Azure, worldwide) and traffic routing (Cloudflare); data is stored in the United States. The full list is at https://trust.anthropic.com/subprocessors.

If the entrusted work or a processor changes, the operator will publish the change in this policy without delay.

Review note

Before the paragraph above is final, each processor's terms and data processing addendum must be mapped to the seven items in Article 26(1) of the Act and Article 28(1) of the Enforcement Decree, and the result recorded. Sub-processing by the hosting provider and the store provider is not listed yet; check the list each publishes and add it. If server logs (including the processing records of AI requests) are sent to a logging service other than the hosting provider, that company must be added to this table and to the overseas-transfer tables too.

10.Overseas transfer of personal information

To provide the service, the operator entrusts the processing of personal information to recipients outside Korea as set out below, and gives the following notice under Article 28-8(2) of the Personal Information Protection Act.

AI service provider: Anthropic, PBC

Legal basis for the transfer

Personal Information Protection Act, Article 28-8(1)3(a): entrusted processing needed to perform the contract with you, with the items of paragraph 2 of that article published in this policy.

For both AI interpretation of the style reports and the Instagram report, your consent to the overseas transfer under Article 28-8(1)1 is obtained as well, before anything is sent.

Type of transfer

Entrusted processing of personal information

Personal information transferred

AI interpretation of the style reports: the kind and version of the report, the display language, each scale's name and score (with its range), percentiles and bands, how many questions you answered, the name and one-line description of your type, and the notes on response quality. Your answers to individual questions, the time you finished the test and the derived numbers the report uses to draw its charts are not transferred.

Instagram report, coding step: downscaled images and captions of the posts you chose from your own account, the post numbers and the report language. Other accounts' names (@) in your captions are masked before the transfer.

Instagram report, reading step: evidence in text form (for each post its number, date, kind, the neutral one-line description and categories, and the start of the caption), counts worked out on your device, and the interpretation text produced from them.

Instagram report, first look: downscaled images (up to 11) and captions you chose from your own account, the image numbers and the report language. Other accounts' names (@) in captions are masked before the transfer, and no dates or account name are transferred.

Human.ai has no member accounts, so no registration details (name, email address, member identifier) are transferred.

Destination country

United States. The operator pins every request to the United States so that AI inference runs only there, and records the inference region each response reports as a check.

When and how it is transferred

Each time you ask for an AI interpretation or an Instagram report, over an encrypted connection (TLS).

Recipient (name and contact)

Anthropic, PBC (contact: privacy@anthropic.com)

Recipient's purpose of use

Generating the AI interpretation text, and analysing post images and captions for it (carrying out the entrusted work)

Recipient's retention and use period

Deleted within 30 days of receipt or generation. Inputs and outputs flagged as a Usage Policy violation may be kept for up to 2 years, and safety classification scores for up to 7 years.

How to refuse the transfer, and the effect of refusing

AI interpretation of the style reports: nothing is transferred unless you consent and ask for one. After consenting you can withdraw in Settings. You can still use the report with the built-in interpretation.

Instagram report: nothing is transferred unless you consent to the overseas transfer. Consent is asked afresh for every reading. Without it the Instagram report is unavailable; every other report still works.

Features of the AI service provider that the operator uses

Pinned inference region

Every request names the United States as its inference region, and the region each response reports is written to the server's processing record as a check.

Prompt cache

The text evidence that is read several times within one reading (the coded table of your posts), and instructions that are the same for everyone, are placed in the AI service provider's short-lived cache. According to the provider's documentation the cache lasts 5 minutes by default, is extended by that much each time it is read, and is not shared with other organisations. Pictures and the scores of the style reports are not marked for caching.

Fallback for a declined request

If the AI service provider's safeguards decline a request, a feature is switched on that has another model from the same provider process the same request once more. The content does not go to any other company, and the report shows the name of the model that actually wrote the interpretation.

Features not used

Features that keep content beyond the handling of a request, such as file storage, batch processing or sending feedback, are not used.

Hosting provider: Vercel Inc.

Legal basis for the transfer

Personal Information Protection Act, Article 28-8(1)3(a): entrusted processing and storage needed to perform the contract with you, with the items of paragraph 2 of that article published in this policy.

Type of transfer

Entrusted processing and storage of personal information

Personal information transferred

Access logs (IP address, time of access, the address requested, browser and device details) and processing records of AI requests (without content)

Content held in server memory while an AI interpretation, Instagram report (including a deep reading's purchase details) or Instagram Login request is processed. It is not stored.

Destination country

United States (primary processing facilities). May be processed in other countries where Vercel and its sub-processors operate

When and how it is transferred

Each time you open the service or send it a request, over an encrypted connection (TLS).

Recipient (name and contact)

Vercel Inc. (contact: privacy@vercel.com)

Recipient's purpose of use

Running the service's servers and keeping access logs and server logs

Recipient's retention and use period

30 days

How to refuse the transfer, and the effect of refusing

You can refuse by not connecting to the service. The service is then unavailable.

Rate-limit and purchase-record store provider: Upstash, Inc.

Legal basis for the transfer

Personal Information Protection Act, Article 28-8(1)3(a): entrusted processing and storage needed to perform the contract with you, with the items of paragraph 2 of that article published in this policy.

Type of transfer

Entrusted processing and storage of personal information

Personal information transferred

A one-way hash of your IP address, made with a secret value, and a count of requests; and the whole service's AI requests per day

If you bought a deep reading in the app: a one-way hash of the transaction number, with how many times the purchase has been used and when it first produced a reading

Destination country

United States

When and how it is transferred

Each time you ask for an AI interpretation, an Instagram report or an Instagram Login connection, and each time the app checks which features are available, over an encrypted connection (TLS).

Recipient (name and contact)

Upstash, Inc. (contact: privacy@upstash.com)

Recipient's purpose of use

Temporary storage for rate limiting, and keeping purchase use records

Recipient's retention and use period

Rate-limit counters are deleted automatically once the length of the limit window has passed since they were first written. The window is 1 minute, 10 minutes or one day, depending on the request. Purchase use records are deleted automatically after 400 days.

How to refuse the transfer, and the effect of refusing

You can refuse by not asking for an AI interpretation, an Instagram report or an Instagram Login connection, and by not buying a deep reading. Those features are then unavailable.

For overseas transfers, the operator agrees the protective measures of Article 29-10 of the Enforcement Decree (security measures, and measures for handling complaints and resolving disputes) with the recipient in advance and reflects them in the contract.

Review note

“United States” is accurate only while every AI request is pinned to US inference. The code names the region on every request and records the region each response reports, but this has never been confirmed in production with a real API key, and the region restriction on the Anthropic workspace is something the operator has to set separately. Whether the same region and retention terms apply when a declined request is handled by the fallback model, and how content placed in the prompt cache and any record the provider keeps for the fallback fit within the contractual retention terms, must be confirmed with Anthropic. Anthropic's published note on sub-processors gives the location of its cloud infrastructure as “worldwide” and says data is stored in the United States. Whether “United States” alone is enough as the destination country when inference is pinned there, or whether the sub-processors' countries must be listed too, needs a judgment. The hosting provider has its primary processing facilities in the United States but reserves processing in other countries where it or its sub-processors operate, so whether that statement meets the requirement to name the destination country needs checking as well. The store provider's processing country is the primary region chosen when the database was created, which only the operator can read in its console. If that region is in Korea, delete the store's table and keep it under entrusted processing alone; if replicas were placed in other regions, list those countries too. In the Instagram Login connection the server sends Meta only the authorisation code and token that Meta itself issued; whether that needs to be treated as an overseas transfer should be confirmed too.

Review note

The server is built to process what it receives in memory only, to keep no copy and not to log request content, and automated checks watch part of that. It has never been confirmed in production, however, so after the first deployment the server logs must be read and compared with these tables. Also, for AI interpretation of the style reports the basis for processing is given as consent (Article 15(1)1) while the basis for the overseas transfer is given as both entrusted processing needed to perform the contract (Article 28-8(1)3) and consent (Article 28-8(1)1). Confirm that stating both is right, or whether the contract-necessity basis (Article 15(1)4) recommended in the research brief is the better construction.

11.Use of generative AI

AI interpretation in the service is written by generative artificial intelligence (AI). It is not written by a person and is not individually reviewed by a professional. To provide it, the operator uses an external generative AI service (the Claude API from Anthropic, PBC). The basic reading and the deep reading of the Instagram report use different models from the same provider, and what is passed on to the AI service provider is the same for both. The first look uses the same model as the deep reading and reads the chosen images and captions in one request.

Generative AI at a glance

Intended use

Writing interpretation text as reference material for self-understanding, limited to interpreting the scores of the style reports and to the Instagram report. It is not for diagnosis or assessment, and not a basis for decisions such as hiring, admissions or legal judgments.

Reports that do not use AI

The self-checks show text written in advance and selected by scoring rules. Nothing is sent to the AI service for them.

Collection of inputs and outputs

The operator does not store what it receives or the interpretation that is generated. The generated interpretation is saved only on your device.

Use for AI training

The operator does not use your inputs or the generated output to train or improve AI models. The AI service provider's retention and use terms are those stated under Overseas transfer of personal information.

Retention of inputs and outputs

The operator: not stored.

The AI service provider: Deleted within 30 days of receipt or generation. Inputs and outputs flagged as a Usage Policy violation may be kept for up to 2 years, and safety classification scores for up to 7 years.

How to opt out of training

No procedure is needed, because the operator does not use them for training. If you do not want anything sent to the AI service at all, withdraw your consent to AI interpretation in Settings; nothing is sent after that. The Instagram report asks for consent afresh at every reading, so nothing is sent unless you agree.

Labelling of output

Interpretation written by AI is labelled as AI-generated.

Checking of output

Text written by AI is checked again against fixed rules before it is shown. The rules are built to remove statements that are not tied to a score or to post evidence, and statements on subjects the reports do not cover. Rules do not catch every error.

  • An AI interpretation can be wrong or may not fit you.
  • Before sending posts for an Instagram report, please leave out any post that contains another person's personal information or sensitive content.
  • Please report an interpretation that is inappropriate or wrong to contact@craving.win.

12.Sensitive information and whether it can become public

  • Answers and results of the self-checks (your answers to questions about mood and attention, records of the attention tasks, drawings and scores) may count as information about health. They are calculated and stored only on your device and are not sent to the operator's server.
  • The posts you choose for an Instagram report can reveal things about health, beliefs, political views, sex life and similar matters. Under Article 23(1)1 of the Personal Information Protection Act, the operator processes them only with a consent that is separate from consent to any other processing.
  • The Instagram report is designed not to infer health or mental-health conditions, sexual orientation, race or ethnicity, religion, political views, income, intelligence, attachment style, anything about minors, or anything about other people who appear in your photos.
  • Posts in which a minor, health, sex, mourning, alcohol or smoking, or words about self-harm are detected are not used as evidence for any interpretation, and posts about religion or politics are left out of interpretations about values. Of these, posts that show or name a minor, show a medical setting or carry words about self-harm are left out of every figure in the report, except that their posting dates count toward posting rhythm. To be detected at all, however, such posts are sent to the AI service at the coding step.
  • A first look started directly, without coding (when you have fewer than 12 posts that can be sent), has no step that detects anything post by post, so the per-post exclusions above do not apply to it. Instead the AI is instructed not to write about health, the body or looks, age, relationships, beliefs, political views, origin, money or ability, or about anyone other than the account owner or about a minor, and before anything is shown its text goes through the same wording checks as a reading, with any statement that fails removed.
  • A first look offered after coding, because too few posts could be read, leaves out the posts the coding step kept from every interpretation (those above in which a minor, health, sex, mourning, alcohol or smoking, words about self-harm and the like were detected). If words about self-harm were detected, no first look is offered, and help contacts are shown instead.

Whether sensitive information can become public, and how to keep it private

Reports are stored only on your device and the operator does not publish them. Content reaches other people only if you use a share function yourself, and then only what you choose. If you do not share, only you can see your reports.

Review note

The self-checks are processed on the device only. A judgment is needed on whether Article 23 separate consent is still required and whether the operator “processes” that data as a controller at all. The research brief recommended separate consent on the assumption of server-side storage. The separate sensitive-information consent for the Instagram report is taken at every reading and not stored, so the operator holds no record that it was given.

13.Security measures

The operator takes the following measures to process personal information safely.

  • Nothing stored: the server handles content received for AI interpretation and the Instagram report in memory only and does not store it. There is no database of user records or reports.
  • Encryption in transit: connections between your device and the server, and between the server and the AI service, are encrypted (TLS).
  • No logging of request content: the scores, images, text and tokens you send are not written to server logs. For AI requests only the step, the outcome, the model, the time taken, token counts, the inference region and counts are recorded, and any value that is not a number or a short identifier-like word is removed before it is written. When an error occurs, only diagnostic details such as the kind of error and its code are recorded, and values that could carry request content, such as the error message, are not.
  • Minimised before sending: images for the Instagram report are sent only after being downscaled and stripped of capture details and other metadata. If an image the server receives still carries metadata, the server re-encodes it to remove that before passing it to the AI service. Other accounts' names (@) in captions are masked on your device and once more on the server.
  • IP address minimised for rate limiting: rate limiting uses a one-way hash in place of the IP address itself. Access logs do contain the IP address.
  • No Instagram token stored: the access token received through the Instagram Login connection is not stored on the server. It is encrypted and left with your browser only, and is no longer accepted after 30 minutes.
  • Purchase records minimised: once a deep reading purchase has been confirmed, only a one-way hash of the transaction number and a record of its use are kept; the transaction record and purchase token the app sends are not stored.
  • Request limits: the number and size of requests are limited, and the total number of AI requests handled in a day is capped, to protect the server.
  • AI inference region pinned and checked: every AI request names the United States as its inference region, the region each response reports is recorded, and a warning is logged if any other region appears.
  • Kept out of device backups: the mobile apps are set to keep the storage that holds your reports out of device backups.

Review note

Add organisational measures (an internal management plan, access control and authentication for the operations and cloud consoles, review of access logs and so on) only as actually practised. Do not list any certification that is not held.

14.Pseudonymised information

Not applicable. The operator does not process pseudonymised information.

15.Cookies and similar technologies, and how to refuse them

The service uses only the two cookies below. Neither contains anything that identifies you.

Language cookie

Name

humanai-locale

Purpose

Remembers the language you chose (Korean or English) so your next visit opens in it.

Value stored

A language code (ko or en)

Kept for

1 year

When it is set

When you change the language yourself

Instagram Login connection cookie

Name

__Host-humanai-ig-oauth

Purpose

Makes sure an Instagram Login connection can only be completed in the same browser that started it (a security measure).

Value stored

A random value

Kept for

10 minutes at most. It is deleted as soon as you return from Instagram.

When it is set

Only when you start an Instagram Login connection

Where it is sent

While it exists (10 minutes at most) it travels with requests to the service. It is stored and sent only over an encrypted (HTTPS) connection, and scripts on the page cannot read it.

Apart from cookies, browser storage is used as follows. Local storage holds your appearance (theme) setting and a single marker of whether this device has any saved report or test in progress (used to choose the opening screen). When you open Instagram's export page from the Instagram feed reading, the time you did so is kept as well, so that you can go straight to choosing the file when you come back (it says nothing about the account or the file, and it is removed once a file is read or after 7 days). Your answers, reports, settings, the records of the adult check and of consent, and the temporary store of post categories are kept in IndexedDB. Session storage is not used. In the mobile app, the deep reading purchase details are kept in the app's own preferences storage (What is stored only on your device). What is sent when you ask for an AI interpretation or an Instagram report (a deep reading's purchase proof included) is only what is listed under What is processed, and on what legal basis; none of the rest of this is sent to the server.

To refuse cookies, block or delete them in your browser settings. The service still works without them, although your language choice may not be remembered and an Instagram Login connection may not complete. “Delete everything on this device” in Settings also removes the language cookie.

The operator uses no advertising or analytics trackers. It does not collect online behavioural information and does not allow third parties to collect it.

To run the service, the operator's server counts how often a few things happen each day: arrivals at the website (a first page load coming from elsewhere), openings of the Instagram report's screen, finished readings and AI interpretations, deep-reading purchases and refund notices, feed readings that failed on the server, and requests turned away by the daily cap on free use. The counts are split only by app (per operating system) or web and by payment environment (real or test). They are plain numbers that distinguish no one: no IP address, device or browser identifier, transaction number or request content is used for them. They are kept in Upstash, Inc. and deleted automatically after 40 days. The operator receives the counts once a day, and is told straight away when a reading or AI interpretation is finished, a purchase completes, a feed reading fails on the server, a refund notice goes out or a daily cap is reached, and when a day's count reaches a set step such as 10, 50 or 100. These notices go to the Telegram messaging service and hold nothing beyond these numbers and categories.

16.Your rights and how to exercise them

As the data subject, you may at any time ask the operator for access to your personal information, for its correction or deletion, or for processing to be suspended, and you may withdraw consent.

Information stored on your device

Your answers and reports exist only on your device, so you can view and delete them yourself.

  • Access: open your reports in the library, or take a copy with “Export my data” in Settings.
  • Deletion: use “Delete everything on this device” in Settings.
  • Withdrawing consent: withdraw your consent to AI interpretation in Settings. It takes effect at once, and nothing is sent after that. Consents for the Instagram report are not stored, and a reading in progress can be cancelled on its own screen.

Information the operator processes

Requests about information the operator processes, such as access logs, can be made in writing, by email or by phone. Email: contact@craving.win. Phone: 010-9027-0729

  • A request for access is answered within 10 days.
  • On a request for correction or deletion, the operator takes the necessary steps without delay and tells you the result.
  • On a request to suspend processing, the operator suspends it without delay and tells you the result within 10 days.
  • If you ask about personal information that was collected from someone other than you, the operator tells you within 3 days where it came from, why it is processed, and that you have the right to ask for processing to be suspended or to withdraw consent.

Because there are no accounts, handling a request may need information that shows the record is yours (for example the time of access and the IP address used, or, for a purchase use record, the app store you bought from and the order number). Only the minimum needed for that check is requested.

You may also act through a legal representative or an authorised agent, who must provide a document showing their authority.

The operator keeps the method and procedure for exercising rights no harder than the method and procedure by which personal information is collected.

If you believe someone else submitted your posts, see Opt-out and privacy requests for how to make a request.

Review note

When a request for access or deletion concerns the inputs and outputs the AI service provider holds during its retention period, confirm from the contract and the actual procedure whether the operator can identify the request concerned and have the processor act on it. There are no accounts and no per-request identifier is kept, so state here what is possible and how once that is settled.

17.Automated decisions

Not applicable. A report is reference information given to you. The operator does not use a fully automated system to process personal information in order to make decisions that significantly affect your rights or duties.

18.Privacy officer

The operator has designated the privacy officer below to take overall responsibility for the processing of personal information and to handle complaints and remedies relating to it.

Privacy officer

Name or role

송현섭 (대표자)

Contact

contact@craving.win, 010-9027-0729

Privacy enquiries

contact@craving.win

Operator

Company

크레이빙(CRAVING)

Representative

송현섭

Address

서울특별시 성북구 안암로 145, 경영본관동 2층 224호(안암동5가, 고려대학교안암캠퍼스)

Phone

010-9027-0729

You can take any privacy question, complaint or request for a remedy that arises from using the service to the privacy officer, and the operator will answer and act without delay.

19.Where to seek a remedy

To seek a remedy for an infringement of your personal information, you can apply to the bodies below in Korea for dispute resolution or advice.

BodyPhoneWebsite
Personal Information Dispute Mediation Committee (개인정보분쟁조정위원회)1833-6972 (no area code)www.kopico.go.kr
Personal Information Infringement Report Center (개인정보침해신고센터)118 (no area code)privacy.kisa.or.kr
Korean National Police Agency cybercrime reporting system (경찰청 사이버범죄신고시스템)182 (no area code)ecrm.police.go.kr

20.Changes to this policy

This privacy policy applies from 1 November 2026.

If content is added, removed or revised, the operator will announce the change in the service before it takes effect and publish the previous policy alongside it so the two can be compared.

Previous privacy policies: none (first version)