OpenAI backs EU transparency code for AI content

Whenever a shocking image goes viral, the knee-jerk reaction is almost always the same: did an AI make this?

To keep digital trust from unraveling, the European Commission just published its Code of Practice on Transparency of AI-Generated Content, a core milestone under the EU AI Act. OpenAI quickly followed up with a public pledge of support, detailing how it tracks and marks synthetic media.

This is not their first time playing ball with Brussels. Back in 2025, OpenAI was the first U.S. tech company to sign the EU's General-Purpose AI Code of Practice. Now, after helping shape this new transparency framework alongside hundreds of stakeholders, it wants to prove its stack is ready for real scrutiny.

Belts and suspenders for synthetic pixels

Marking AI media sounds straightforward until you realize how easily digital signals break.

OpenAI started stamping C2PA metadata onto DALL-E 3 images back in 2024. The problem with metadata is that it is notoriously fragile. A quick screenshot, a file reformat, or a simple upload to a social app that strips EXIF data will wipe it clean.

To stop that leak, OpenAI is deploying a layered approach across ChatGPT, Codex, and its API. Images now carry both C2PA credentials and SynthID watermarks.

Metadata handles rich provenance context, while the invisible watermark stays resilient against transformations. If you want to check an image yourself, OpenAI has a public verification portal live at openai.com/verify.

An admission that watermarks have limits

Refreshing honesty here: OpenAI is explicitly not pitching this as a silver bullet.

The company acknowledged that provenance tech is still very much in its infancy. Watermarks degrade after heavy editing, metadata gets stripped, and labels only work if platforms actually bother to display them.

Because forensic signals fail, OpenAI argues provenance has to be paired with abuse classifiers, likeness protection, and election safeguards. It also nudged European regulators to keep rule enforcement grounded in what current technology can realistically do.

Why it matters

For everyday users and fact-checkers, having a free tool at openai.com/verify is handy for checking suspicious viral images. But it is not a foolproof shield. Anyone determined to deceive can still crop, filter, or re-render an image enough to shake off detection.

For the AI sector, this is a calculated regulatory play. Europe's rules carry massive fines, and OpenAI is aggressively positioning itself as the most cooperative foreign player in the room. By voluntarily baking both C2PA and SynthID into its pipeline, it sets an expensive compliance baseline that rivals will be forced to match.

We think combining C2PA with Google's SynthID is a pragmatic move that bridges fragmented industry camps. However, the real bottleneck is interoperability. If every lab launches its own siloed verification portal, expecting the public to hop between different checkers to verify an image simply will not scale. Until browsers and social feeds verify these credentials natively, provenance remains an imperfect defensive layer.