‘AI Image’ Reported as Someone’s “Own Photo”… Police: “Identical Photo,” ‘Search, Seizure and Inspection Warrant’ Executed in the U.S. via International Cooperation

This record does not pass judgment on any individual or institution in advance. It sets out, in chronological order, what has been confirmed at each stage of this case and what has not. People often say ‘AI photo,’ but to tell it apart from real photos, this record says ‘AI image.’

On October 4, 2026 at 9:55 p.m., the investigator in charge asked by text message for materials related to my ChatGPT image generation. Confirmed fact The request in the message reads:

“Regarding what you told me earlier about generating images with ChatGPT, please send me the related materials” (일전에 연락주신대로 챗지피티 이미지 생성하셨다고 했는데 관련 자료 보내주세요)

This request follows the October 2 call, in which the investigator asked me to send screenshots. The October 2 call is covered below, in the part where the investigator asked me to send materials. As of October 4, I have not heard any investigation result.

On October 5, I submitted the requested image-generation materials by email.

On October 6, I asked OpenAI’s official account (@OpenAI) and CEO Sam Altman’s account (@sama) a public question on X, and I am now waiting for a reply. The questions and the posts are in What I Asked OpenAI below.

The investigator in charge explained it this way: the reporter came to the police station in person, showed their face, account and posts, and filed the report, and the investigator looked at the reporter's face and said the person in the photo was indeed them. The investigator said the reporter's photo and the image I used as my profile picture were identical, that a warrant was issued, and that it was executed in the United States through international cooperation.

What the investigator said about this image in the first call is as follows (the original is below).

When I heard the word “theft” (도용) in the first call, and “search, seizure and inspection warrant” (압수수색 검증영장) and “international cooperation” (국제 공조) in the second, I was taken aback.

This can be unfair to both sides. If the investigator's account is right, the reporter may be a victim whose photos have been stolen in many places. On the other hand, someone who made a human face with AI and used it may be reported on suspicion of ‘photo theft’ because that face was judged identical to a real person's photo, and may then go through having their personal information handed over via a warrant and international cooperation.

When I asked whether the reporter had submitted a photo, and whether that photo was identical to the image I used as my profile, the investigator answered like this (the original is below).

A little later, when I asked again whether the two images were identical, the investigator first said “I can't tell whether it was made and brought in, or ChatGPT took it from somewhere…” (만들어 온 건지 챗gpt가 따온 건지는 알 수가…), and the rest was cut off because I kept talking. Then the investigator said “Yes, identical. I told you the same thing back then too, didn't I” (예 동일하다고요. 그때도 똑같이 말씀드렸잖아요) (the original is below).

As my records show, I entered only two text prompts and no reference photo. If the investigator's account and my records are both correct, this becomes a question that OpenAI, the maker of ChatGPT, has to answer. And if a warrant did come out as the investigator described, what was ‘confirmed’ at that time?

Making a human face with AI has become something anyone does. I am writing this record so that people who may go through the same thing can see what to keep and how to respond, and to leave questions about the procedures that involve AI images and about ChatGPT, which made the image.

This record is not meant to blame the reporter or the investigator. If the investigator is right, the reporter is someone whose photo has been stolen on many accounts and who may be a victim. In the two calls the investigator explained the procedure, and in the first call said “…please hold on to your proof materials well” (…그 입증 자료 잘 가지고 계시고요). Information that could identify the police station or the investigator is not included in this record.

This question is one of the biggest reasons I am recording and publishing this case myself instead of leaving it as a mere episode. I set out what I know and what I do not, to help anyone going through something similar.

On the October 2 call, the investigator said, “But you don't have any status yet, I'm telling you.” (근데 무슨 신분이 아직 있으신 게 아니라고요.)

What I know: The investigator explained that my account's user information was obtained from Meta through a search, seizure and inspection warrant and international cooperation (October 2 call).

What I do not know: I could not confirm where or for how long that information is kept, or when it is deleted.

Notice I received: I did not find any such notification in the notifications of the Instagram app or the Threads app. I may have missed one because I am not used to the apps. I could not find one in my email either.

I could not confirm whether these provisions apply to this case as they stand.

On July 10, 2026, in the free version of ChatGPT (not a paid plan), I asked for an image of a person using text. I did not provide any reference photo to make the image. Confirmed fact

I felt the first result looked AI-made, so I asked once more. Confirmed fact

I used this second result as my profile picture. The image that was reported is this one, and the pixelated image at the very top is exactly this image. Confirmed fact

This sub-account was one I made to try things out, and that is also why I used an AI image as the profile picture. Looking at it now, it had been left unattended with only a few posts and comments.

I do not know who the reporter is. I have never seen their photo, face, name or contact information, I have no way of knowing them, and I have no intention of looking for them.

I am not publishing the original of the second result. If, as the investigator said, it is identical to the photo of a real person, I judged that publishing the original could cause secondary harm to someone who is already struggling with photo theft. If grounds for a judgment come out through the police or OpenAI, I will decide again then.

That I did not include a reference photo can be confirmed from screenshots of the conversation and the library (ChatGPT's image storage).

I am keeping the ChatGPT conversation records and the original files, all unmodified.

I left this sub-account as it was even after the first contact on September 30 and the second call on October 2, to keep it as evidence. This case is still in progress, and the investigator said this image is identical to the photo of a person who has a real account. If the account is found through an image search or the like, that person could be identified, so to avoid secondary harm to the person who filed the report, on October 4 I set the Threads and Instagram sub-accounts to private and took down the profile image. The screens from before I took it down, the original files, and the posts and replies are all kept as they were, without being deleted or edited.

Given that the investigator said the two images were identical, there are four cases to consider. At this point I cannot tell which one is true.

To be fair, I set down the reasons the report and the investigation can be seen as justified, together with the call transcripts.

Even if I accept all of these reasons, how an image made from two lines of text came to be ‘identical’ to the photo of a real person has still not been explained.

There are two ways to report photo theft. You can report an impersonating account or a privacy violation to the platform where the photo was posted, or, if you want an investigation, you can report to the police cyber crime reporting system. Whichever you choose, first record the post and the account.

The last two rows are channels that come with conditions. The Korea Internet & Security Agency's Personal Information Infringement Report Center states that it serves people whose rights have been infringed by a personal information controller (a business or service), so I do not treat it as a general channel for photo theft between individuals. I could not confirm from an official source where portrait rights infringement begins and ends. If you need a judgment, I recommend getting legal advice.

According to the investigator, the reporter in this case also saw that the photo had been moved to ‘many accounts’ and went to the police station in person to report it. The investigator said in the first call that the photo had gone up in many places, and also said “I don't know whether they're all people like you” (다 선생님 같은지는 모르겠어요) (September 30). If the reported accounts may include one that used an image made by AI, records need to be kept together so that neither the reporting side nor the reported side is wronged.

If you received a ‘photo theft’ report over an image made by AI, the place to start is to leave the conversation records and the original files that made that image undeleted and as they are. In ChatGPT, a data export cannot bring back a conversation that has already been deleted. I made mine from two lines of text with no reference photo, and I have not deleted those records.

I made the call asking about the progress, two days after the first call, myself. The results of my own test are in the frequently asked questions. How to check whether an image was made by AI follows right below.

Checking an AI image starts by uploading the original file you first downloaded to OpenAI's generated-content verification tool and seeing whether C2PA metadata or a SynthID watermark is detected. This tool looks for the OpenAI signals placed in images and audio made with ChatGPT, the OpenAI API and Codex, and it does not tell you who made something and why, or whether it was made with another AI service. The Content Credentials verification tool shows the C2PA record in a file and its change history, and does not tell you the origin of a file with no record. SynthID is checked by uploading to Gemini to see whether something was made or edited with Google AI, and I could not confirm whether it catches images made with a tool that is not Google AI (for example, OpenAI).

When I uploaded the reported image 2 to OpenAI's generated-content verification page, it showed ‘Generated with OpenAI tools,’ and the creation time was shown as July 10, 2026.