The European Commission's AI Office began full enforcement of the AI Act's general-purpose model obligations on Monday, dispatching formal documentation requests to six providers including OpenAI, Anthropic, Google DeepMind, Meta, Mistral, and Cohere.
Maximum penalties under the regime are €35m or 7% of global annual turnover, whichever is higher — a structure deliberately mirrored on the GDPR. Officials stress that fines are a last resort and that early enforcement will focus on transparency obligations.
What providers must disclose
The requests cover training data summaries, downstream-use risk assessments, and copyright-compliance attestations. Providers have 60 days to respond.
The political question is whether the Office can credibly enforce against US-headquartered labs without cooperation from American regulators. The answer, for now, appears to be yes: each of the six firms has confirmed it will comply.