MatIntellect

👀 On May 19 OpenAI quietly launched a tool that changes the rules for AI images. Upload any picture - and a few seconds later you know whether ChatGPT drew it or not

What is Verify:
The tool is public and free. You open openai.com/verify, upload a file - and get a report: OpenAI markers or not. Starting May 19, 2026, every image created via ChatGPT, the OpenAI API and Codex automatically gets two invisible signals embedded in the image data. One is cryptographic in the metadata, the other is right in the pixels. No settings needed on the user's side

Two technologies under the hood:
📌 C2PA Content Credentials - cryptographically signed metadata with the source, the tool used to create it and a timestamp. An open standard: backed by Adobe, Microsoft, Google, Reuters, BBC and several hundred more organizations. Weakness: metadata can be stripped programmatically
📌 SynthID from Google DeepMind - invisible watermarks embedded right into the pixels. They survive screenshots, compression, cropping and format changes. That's exactly why the two technologies work as a pair: where one fails, the other takes over

OpenAI honestly admit it: errors are rare but possible. The main limitation today - Verify only sees images from OpenAI products. Upload a picture from Midjourney or Stable Diffusion - you'll get an empty result, and that doesn't mean it's not AI-made

What happens next with the ecosystem:
⚡️ Google Search and Chrome - getting ready to build native SynthID watermark detection right into the browser and search results
⚡️ Adobe via CAI - has been pushing C2PA as a global industry standard for years through a coalition with media, tech and camera makers
⚡️ The road to full coverage - only once the whole market adopts C2PA will detection stop being "just an OpenAI thing" and become a real infrastructure layer

💼 Verify is available for free at openai.com/verify - no sign-up, no limits

How I work with AI images myself

I don't hide that AI makes my images. Since April I've made 296 AI covers, and the designer agent on my team has drawn 94 images in three months, each only after my yes. At the same time, content factory posts use stock photos by default: a real photo where credibility matters, AI where you need an idea or a cover

Social platforms already apply labels. According to Meta, the "AI info" label appears when a file carries industry-standard AI indicators that generators leave in it, or when the author discloses that the content was made with AI. Since September 2024, content that was only edited with AI tools has the label moved to the post's menu, while fully generated content still shows it openly

What this means for a business

Hiding where an image came from is pointless: more and more services read the indicators in the file. It's far more useful to decide in advance where you need real photography and where an AI image honestly does its job. An expert's face, a product, the result of your work are better shot for real. Covers, diagrams and illustrations of an idea can safely go to AI

I treat video the same way. My website has 40 explainer videos, and 22 clips have gone through automatic subtitles: I film them myself, and AI takes the tedious part of editing. My face and voice are on screen, so the label question doesn't even come up, and editing time dropped from about half an hour to a couple of minutes per video

💭 OpenAI made Verify a free public tool - they're not monetizing it, they're building trust infrastructure. The weak spot is obvious: for now it only works for pictures from OpenAI itself. But the direction is clear - once Google builds detection right into Search, images without metadata will start raising suspicion automatically. The logic will flip: not "found a marker - so it's AI", but "no marker - so something's off here" 👀

Share:

No comments yet

Leave a Comment

Fields marked with an asterisk (*) are required