SpineCard

THE AUTHENTICITY OF THE ALLEGED PLAION DOCUMENT HAS FINALLY BEEN CLARIFIED The file presented as an internal PLAION document announcing the cancellation of the Neo Geo AES+ can now be technically traced. The analysis shows that it was generated with ChatGPT / gpt-image, then transmitted under a filename that almost exactly reproduces that of an older official PLAION asset announcing the console’s delay. All the evidence converge towards the fabrication of a fake document designed to look like a genuine PLAION leak. One essential clarification: @SpineCard sent me the exact file he himself had received as an email attachment, not a screenshot or an image recovered from X. That original file is the one that was used for the analysis. DIRECT AI ANALYSIS — A FALSE NEGATIVE The first method was simply to upload the image to ChatGPT or Gemini and ask whether they detected any traces of AI use. The result could be negative: no clear indication of AI, no C2PA manifest detected. That result was nevertheless incorrect. The reason is that an image uploaded directly to an AI system can be re-encoded or transformed during the upload process. The analysis may therefore no longer be performed on the exact same bytes as the original file. One detail illustrates this: ChatGPT described the copy it was analysing as RGBA, while the original PNG extracted from the file sent by SpineCard is RGB. Having an AI directly analyse an image was therefore not a sufficiently reliable method for examining its metadata and C2PA structure. The original file had to be analysed locally. EXIFTOOL — LOCAL ANALYSIS OF THE ORIGINAL FILE A local analysis of the original PNG with ExifTool revealed what had been invisible during the direct ChatGPT analysis. The file contains, among other things: c2pa.created, c2pa.converted, c2pa.watermarked.unbound, ChatGPT, gpt-image, OpenAI Media Service API and trainedAlgorithmicMedia. Most importantly, ExifTool reveals the timestamp recorded in the manifest at the time of creation: 25 September 2026 — 20:17:30 UTC, i.e. 22:17:30 local time in Spain / Central Europe. However, the email presenting this image as a genuine PLAION document was sent to SpineCard at 22:29. The image had therefore been generated approximately 11 minutes and 30 seconds before the email was sent. This is also the point at which another particularly important piece of information appears: the manifest associated with the OpenAI generation gives the file the title image.png. CONTENT CREDENTIALS — ONLINE VERIFICATION The original file was then submitted to the Content Credentials verification service. The service identifies OpenAI OpCo, LLC, OpenAI Media Service, a creation operation and a format conversion operation. It explicitly states that the image was generated using an AI tool. So this is no longer simply an interpretation of ExifTool: the C2PA credential embedded in the file is also recognised by the Content Credentials verification tool. C2PATOOL — VALIDATION OF THE MANIFEST AND OPENAI SIGNATURE The analysis was then taken further with c2patool, the dedicated tool for inspecting and validating C2PA manifests. It finds the same manifest, including OpenAI Media Service API, ChatGPT, gpt-image, the title image.png, and the same creation timestamp of 25 September at 20:17:30 UTC. The certificate chain identifies OpenAI Media Service and OpenAI OpCo, LLC. C2patool validates, among other things, the claim signature, the assertion hashes, the data hash, the signing certificate within the C2PA trust chain, and the fact that the certificate had not been revoked. At this stage, it can therefore be technically established that the image was created by OpenAI’s ChatGPT / gpt-image pipeline, at the time recorded in its manifest. AI GENERATION ALONE DOES NOT PROVE THE DOCUMENT IS FAKE But this still does not, by itself, prove that the document is fake. PLAION could perfectly well have decided to use an AI tool to produce a visual for its communications. Such a scenario would not even be implausible within the SNK environment: there has already been the apparent use of AI-generated imagery in an official Kim trailer for COTW, as well as the image used to present the mosaic project for the Metal Slug 30th Anniversary. THE FILENAME COMPARISON This is where the comparison becomes particularly interesting. In the C2PA manifest corresponding to its generation by OpenAI, the file appears under the title image.png. But the file received by SpineCard in the email was named 1HRsndEsW8AAx_5a.png. Yet there is a genuine earlier PLAION asset named HRsndEsW8AAx_5a.jpg, corresponding to the announcement of the console’s delay on PLAION’s Spanish account. The match is almost exact: Official PLAION file: HRsndEsW8AAx_5a.jpg File presented as the leak: 1HRsndEsW8AAx_5a.png The name of the older official asset is therefore reproduced character for character, with only a 1 added at the beginning and the extension .jpg replaced with .png. At the same time, thanks to the C2PA manifest, we know that when the image passed through the OpenAI pipeline it was associated with the generic title image.png. It can therefore be established that the file was renamed between its generation and its transmission by email. The file itself does not allow us to determine who renamed it, nor to directly prove the intention of the person who did so. But the result is objective: after being generated by ChatGPT, the image ended up with a filename almost identical to that of a genuine earlier PLAION asset, giving it the appearance of a file originating from the same publishing environment. CONCLUSION Taken separately, none of the initial elements was enough to reach a definitive conclusion. The fact that an image was created with ChatGPT / gpt-image does not, by itself, demonstrate that it is fraudulent: a company can perfectly well use artificial intelligence to produce a communication visual. But here the elements add up: the exact file received by SpineCard was analysed, its C2PA manifest establishes generation by ChatGPT / gpt-image, that generation is timestamped at 22:17:30, the email presenting the image as a genuine PLAION leak was sent at 22:29 — approximately 11 minutes and 30 seconds later — the manifest initially associates the image with the generic title image.png, the file is then renamed 1HRsndEsW8AAx_5a.png before being sent, and that new name almost exactly reproduces HRsndEsW8AAx_5a.jpg, the name of a genuine earlier PLAION file. So it is not the use of AI, taken in isolation, that allows us to reach a conclusion. It is the combination of the OpenAI provenance, the extremely short timeline, and above all the renaming that reproduces the name of an older official PLAION asset that makes all the evidence converge towards the fabrication of a fake document designed to look like a genuine PLAION leak. #SNK #NeoGeo #PlaiOn