OpenAI published a 31 July post titled "Advancing responsible AI across Europe" S¹, claiming its safety, security, transparency and provenance practices support responsible AI governance across the bloc. The company says this work will continue as the EU AI Act advances S¹. What the post does not include is any specific compliance milestone, audit result or regulatory approval, and that gap is where the real story sits.

My read: This is the first time I've seen OpenAI bundle its safety and provenance practices into a single Europe-facing narrative since it signalled intent to sign the EU Code of Practice for General Purpose AI in July 2025 P². I don't buy that this post is purely informational. It reads like positioning ahead of binding regulation, a way to say "we're already doing the right thing" before anyone forces the issue. The risk is that stated intentions get conflated with actual governance outcomes, and the evidence here is entirely self-reported with no independent corroboration.

What the post actually claims

Three claims, all from OpenAI itself S¹. First, that the company's safety, security, transparency and provenance practices support responsible AI governance in Europe. Second, that this work will continue as the EU AI Act advances. Third, by implication, that OpenAI is already aligned with where European regulation is heading.

None of these claims come with supporting detail. There is no mention of specific EU AI Act provisions or a compliance timeline, and no third-party audit or description of what "provenance" means in practice. The post is a statement of posture, not a compliance document.

This lands three weeks after OpenAI launched its ChatGPT Work agent, which runs across users' apps for hours on end. A company building autonomous agents and a company pledging responsible governance are not necessarily in conflict. But the two posts arriving in the same month tells you where OpenAI's attention is split.

From voluntary code to binding law

OpenAI's relationship with European AI regulation has been building for over a year. In July 2025, the company announced its intention to sign the EU's Code of Practice for General Purpose AI, subject to the code being formally approved by the AI Board during its adequacy assessment P². That code is a voluntary framework for AI developers, a stepping stone before the binding provisions of the EU AI Act take full effect.

The EU AI Act itself is now being turned into machine-readable form by independent developers. A GitHub project by Jeroen Herczeg converts the official EU AI Act Formex XML into a structured, multilingual dataset on Hugging Face P⁴, a sign that the regulatory text is being treated as engineering infrastructure rather than legal prose.

Europe is not waiting for OpenAI

While US companies position themselves around European regulation, Europe is building its own models. TildeOpen-30b-64k, a 30-billion-parameter model developed by Tilde.ai and funded by the European Commission through the EuroHPC JU Large AI Grand Challenge, covers European languages including Albanian, Bosnian, Bulgarian, Croatian and Czech P³. It is a dense decoder-only transformer, the same architecture family as the models OpenAI builds, but trained with public money for public linguistic coverage.

This matters because OpenAI's governance post arrives in a market where European institutions are actively creating alternatives. A company can pledge responsible practices, but if European regulators and funders decide that sovereignty over AI infrastructure is the real goal, corporate safety pledges may not be enough.

What to do about it

For a Brussels-based compliance consultancy that audits AI systems for financial services clients, this post changes almost nothing on the ground. The EU AI Act's requirements are what they are. OpenAI saying it supports responsible governance does not exempt downstream users from their own obligations. A compliance team using an OpenAI model in a regulated workflow still needs to document the model's role, assess bias risks, and maintain human oversight. OpenAI's pledge does not change those obligations.

The practical step this week: check whether your AI vendor's documentation includes provenance information (traceable records of training data sources and model versioning). OpenAI says it has provenance practices S¹, but the post does not say whether those are available to customers. Ask your vendor for the audit trail before a regulator asks you.

What we don't know yet

The post does not specify which provisions of the EU AI Act OpenAI considers itself compliant with, or on what timeline. There is no independent verification of the safety, security, transparency or provenance practices the company claims. No financial investment figures for European operations are mentioned, and no partnerships with specific European governments or regulators are named.

The next signal: the EU AI Board's adequacy assessment of the Code of Practice for General Purpose AI, the same process OpenAI said it was waiting on in July 2025 P². When that assessment concludes, we will know whether the voluntary code OpenAI intends to sign becomes the baseline or gets tightened. We will check OpenAI's compliance claims against it.

Subscribe to see whether the EU's next move validates OpenAI's posture or forces it to go further.

Sources: S1 — Advancing responsible AI across Europe · P2 — The EU Code of Practice and future of AI in Europe | OpenAI · P3 — TildeAI/TildeOpen-30b-64k · Hugging Face · P4 — jeroenherczeg/eu-ai-act-dataset · P5 — ArtifactLinker: Linking Scientific Artifacts for Automatic State-of-th

More from Not A Tech Guy

Generated from an audited evidence pack with primary-source research. Social-media items are discussion signals, not verified facts. Nothing here is financial, legal or medical advice.