Commissioner Henna Virkkunen tells Euractiv the orders cover cybersecurity, safety, and copyright summaries.
The European Commission has issued its first information requests under the AI Act to developers of the most advanced models, Tech Commissioner Henna Virkkunen told Euractiv. The orders cover cybersecurity, safety, and copyright compliance. Virkkunen did not name the companies.
‘All the providers are not covered with this request for information because, with some of them, we have already had very close dialogue during the last months,’ she said. Anthropic, OpenAI, Google, Meta, and Mistral did not immediately say whether they had been contacted, Euractiv reported.
Virkkunen said the Commission wants to know whether the labs have adequate cybersecurity and physical protections around their models. ‘It’s very much about how they are mitigating the risks, to prevent humans or AI from stealing the model,’ she told Euractiv. Officials are also asking how much access outside evaluators get, whether their safety recommendations are followed, and how labs monitor a model once it is public.
Separately, EU AI enforcers asked more than 30 companies to explain how they are meeting European copyright rules. Under the AI Act, labs must publish a summary of the data used to train their models. Euractiv said many have offered thin disclosures, or none. Virkkunen said the Commission went after developers that failed to publish the training-data form or did not answer informal questions about copyright. That is a training-data summary duty, not the law’s separate rules on labelling AI-generated content.
Companies have to answer a formal information request. If they do not, the Commission can send another and, if that one is ignored, fine them. Euractiv noted the fuller toolkit: the Commission can run its own model evaluations, order specific measures, and, as a last resort, levy fines of up to 3 percent of annual turnover.
The first letters are out. The labs on the short list would rather not say if theirs arrived.