OpenAI said it is aligning its safety, security, and transparency work with the European Union AI Act’s General-Purpose AI Code of Practice [1].
This move comes as the EU begins enforcing the AI Act, granting regulators the power to restrict market access for non-compliant models. The stakes are high for AI developers, as the European Commission can now demand inspections of models and levy significant financial penalties [3].
In a press release, OpenAI said it has outlined how its safety and transparency work aligns with the GPAI Code [1]. The company also said it contributed to and endorsed the code [1]. However, reports indicate the company's alignment may be incomplete. While the statement addresses watermarking and safety frameworks, it omits the GPAI Code's chapter on copyright [2].
Enforcement of the EU AI Act begins on a Sunday in 2026 [2]. The timing is critical for the release of new models. Training-data transparency obligations apply to GPT-5 and every model released after Aug. 2, 2026 [2].
Failure to comply with these regulations could be costly. The European Commission can fine providers up to 15 million euros or 3% of their annual turnover [3]. The Act was published in the Official Journal, which triggered a 20-day legal deadline [4].
OpenAI is among several firms, including Anthropic, facing increased scrutiny under these new enforcement powers [3]. The EU's ability to inspect models ensures that companies cannot simply self-certify their safety standards without government oversight [3].
“OpenAI has outlined how it aligns safety, security, and transparency work with the EU AI Act’s GPAI Code.”
The gap between OpenAI's stated alignment and the omission of copyright details suggests a looming conflict between AI labs and EU regulators over training data. Because the EU AI Act mandates transparency for models released after Aug. 2, 2026, OpenAI's handling of copyright will likely determine whether GPT-5 and future iterations can be legally deployed in the European market without risking massive turnover-based fines.
