Is that image AI-made? OpenAI backs the EU's content-transparency Code!

Hey everyone, it's me, Shiichan! Today I've got a slightly grown-up topic about proving what AI made. A world where you can tell whether an image or piece of text online came from AI is going to matter a lot, right?

OpenAI News openai.com

What was announced?

This one comes from OpenAI's News. After the European Commission published its Code of Practice on Transparency of AI-Generated Content, OpenAI announced its support for the Code and the ambition behind it.

The Code is an important step toward implementing the EU AI Act and building a more transparent digital ecosystem. OpenAI says it helped shape the Code alongside hundreds of other stakeholders.

For context, in 2025 OpenAI became the first US company to sign the EU's General-Purpose AI Code of Practice. It's backing this new Code in the same spirit.

Why it matters

As more content is created or edited with AI, it gets harder to know "where did this come from?" and "is it really what it looks like?" That's where provenance comes in.

Provenance signals give people context about a piece of content's source, how it was created or edited, and who signed that information. That helps people interpret content with confidence, makes disinformation campaigns easier to spot, and can support election integrity.

What changes

Transparency isn't something one company can pull off alone - it takes everyone along the value chain. Both the EU AI Act and the Code recognize this, and OpenAI is committing to do its part by complying with the requirements that apply to its relevant products.

OpenAI's view is that clear, workable rules let AI develop responsibly while giving companies the certainty to keep building.

Dive Deep

To make provenance more resilient, OpenAI takes a multi-layered approach that mixes different signals, product safeguards, and collaboration across the ecosystem. Here's the gist:

• C2PA Content Credentials for images: It started adding C2PA metadata to images from DALL·E 3 in ChatGPT and the OpenAI API, and keeps expanding provenance signals across OpenAI-generated images.
• Multiple signals: Images generated with ChatGPT, Codex, and the OpenAI API carry both C2PA metadata and SynthID watermarks. Metadata can hold richer info, while watermarks help keep a signal across contexts.
• A public verification tool: openai.com/verify lets you check whether a supported image contains provenance signals tied to OpenAI-generated images.
• Contributions to open standards: Back in 2024, OpenAI joined the C2PA Steering Committee, a cross-industry effort with software companies, camera makers, media orgs, and online platforms.
• Product safeguards: Policies, classifiers, reporting channels, and enforcement processes that cut down on deceptive uses like likeness misuse and election-related deception.

Honestly though, provenance is still a young field. Metadata can be stripped or lost through uploads and downloads, and broken by things like format changes, resizing, or screenshots. Watermarks can degrade, and labels only help where people actually encounter the content. So improving reliability, robustness, and interoperability still needs a lot of cooperation across the ecosystem.

OpenAI thinks the Code should stay grounded in methods that work in practice and flexible enough to reflect today's limits. It wants to keep working with the AI Office, Member States, and the wider AI community toward a safer, more transparent ecosystem.

Wrap-up

• OpenAI announced support for the European Commission's Code of Practice on Transparency of AI-Generated Content
• The Code is a step toward implementing the EU AI Act, and OpenAI helped shape it
• It's underpinned by provenance tech - a multi-layered mix of C2PA metadata and SynthID watermarks, with openai.com/verify letting anyone check an image's provenance
• But provenance is still developing, and OpenAI is honest that signals can get lost

If you care about spotting AI-generated content, follow EU AI regulation, or build products that deal with content trust, this one's for you.