OpenAI рассказала о практиках безопасности и прозрачности в рамках EU AI Act

OpenAI published a post on its blog about the safety, transparency, and data provenance practices the company applies to support responsible AI governance in Europe. This work will continue as the EU Artificial Intelligence Act (EU AI Act) moves into the practical implementation phase.

What OpenAI's Approach Includes

OpenAI structures its European strategy around four pillars: safety, security, transparency, and content provenance verification.

By safety, the company means risk assessment procedures and model testing before release. Transparency refers to openly informing users about the capabilities and limitations of its products. Provenance verification covers tools that allow distinguishing AI-generated content from human-created content — critically important in the context of combating disinformation, which the EU considers one of its key regulatory priorities.

As

OpenAI's blog states, the company shares "how its safety, security, transparency, and data provenance practices support responsible AI governance in Europe."

When the EU AI Act Takes Effect

The EU AI Act is the world's first comprehensive law regulating AI at the level of a major economic bloc. The European Parliament passed it in March 2024, and it entered into force on August 1, 2024.

Key dates for phased application, according to the European Commission:

• February 2025 — ban on systems with unacceptable risk: manipulative technologies, citizen social scoring
• August 2025 — requirements for general-purpose AI models (GPAI), including OpenAI products: transparency of training parameters, documentation of capabilities
• August 2026 — full requirements for high-risk systems: medicine, education, recruitment
• For GPAI models with training compute exceeding 10²⁵ FLOPs, additional systemic risk assessment requirements apply

Why This Matters for OpenAI in Europe

The European market is one of the key markets for OpenAI: ChatGPT and the company's API are used by millions of users and thousands of corporate clients across EU member states. Publicly demonstrating compliance before requirements become mandatory is a standard strategy for major tech players. Similarly, Google and Microsoft proactively built GDPR-compliant infrastructure starting in 2018.

For OpenAI's models, EU AI Act requirements for GPAI will begin to apply as early as August 2025. Timely preparation reduces the risk of enforcement restrictions from regulators and simplifies due diligence for corporate clients in the EU, who are required to verify the compliance of their AI providers.

What This Means

OpenAI is consistently building an image as a regulatory-responsible partner for European authorities even before most EU AI Act requirements become mandatory. For users and companies in the EU, this is a signal that the company's products will in all likelihood remain available in a regulated mode as regulations tighten in 2025–2027. The real test will come when regulators move to active enforcement.