EU sends first AI Act information requests to OpenAI and Anthropic

Under the AI Act's general-purpose AI regime, the Commission's AI Office can demand documentation from model providers on training data, evaluations and systemic-risk mitigation. Information requests are typically the first formal step before any investigation or fine.

Targeting OpenAI and Anthropic first is notable. Both are expanding in Europe: OpenAI added 200 Dublin roles this week, and Anthropic's frontier models are among those most scrutinized for systemic risk. Anthropic also published research this week warning that small models could self-replicate across consumer devices. That kind of disclosure is exactly what regulators may now ask providers to substantiate.

The same roundup reports researchers showing AI chaining Chrome zero-days in six days, and banks trying to offload roughly $38B of AI data-center debt. Those are separate developments, but they are part of the backdrop driving regulators' urgency.

Caveat: details come from a secondary roundup, and the scope of the requests has not been published. Watch for whether Google, Meta and Mistral receive similar requests, and what deadlines the Commission sets.