PlainAI AI explained plainly

European Union · 2025 · 08 · 02 · Policy · ~2 min read

EU AI Act phase 2 (general-purpose AI rules) took effect

Twelve months after the EU AI Act was published, the rules for general-purpose AI providers kicked in. OpenAI, Anthropic, Google, Meta, Mistral and others now have to publish documentation, manage risks, and cooperate with EU regulators on the models they ship.

What's actually new

  • Documentation requirement. Every general-purpose AI provider must publish technical documentation about their model — capabilities, limits, intended uses.
  • Risk management process. Providers must show how they identified, tested, and reduced risks before release.
  • Stricter rules for 'high-impact' models. Anything above a compute threshold gets extra scrutiny — cybersecurity safeguards, external audits, incident reporting.
  • Mandatory cooperation with deployers and regulators who use or oversee the model.

If you want more

Worth knowing~30s
  • Enforcement in 2025-2026 was light-touch. National authorities were still being set up. Most providers got guidance letters, not fines.
  • Many vendors quietly geo-restricted features rather than fully comply — 'available in the EU' kept asterisks throughout 2025-2026.
  • The high-impact threshold is technical (compute used in training) and will need updating as training methods change.
Who should care~20s

Anyone deploying general-purpose AI in the EU — not just providers. Companies whose AI vendors are now under the rules. Anyone tracking how AI regulation differs across the EU, US, and UK.

What to do about it~20s

Confirm your AI vendor publishes EU-required documentation. Read it once — the disclosures are surprisingly informative about how your model was actually built. Watch for guidance from your national supervisor authority. The next deadline (high-risk system rules) is August 2026.

Honest take~45s

Phase 2 was the EU AI Act getting real for the labs themselves, not just for the people deploying AI. Compliance costs are real — Anthropic, OpenAI, and others spent serious money on EU-specific documentation and audit teams. The deeper effect was forcing the labs to write down what their models actually do, in language a regulator can read. That alone reshapes how the AI industry talks about itself.

Other recent policy updates

Sources

Last verified · 2026 · 05 · 05 · Found a fact wrong? corrections@aguidetocloud.com