The European Commission is requesting information from providers of highly capable AI models about security, independent evaluation and post-release monitoring. Executive Vice-President Henna Virkkunen described the requests already sent in her published October 9 remarks to the AI Scientific Panel.
The AI Office asks about containment that models cannot bypass, traceable computing access and adequate access for independent evaluators. Virkkunen says providers’ replies are arriving.
An official October 12 readout confirms that the office has already requested and received serious incident reports. Further information requests will combine the panel’s and the office’s questions about recent incidents.
The panel’s leadership recommends stronger governance of internal development models, better investigation capabilities and international crisis-prevention cooperation. It also calls for closer scrutiny of training techniques that make models harder to monitor.
In her published October 14 Dublin speech text, Virkkunen reiterates the information requests and calls for AI to remain under human oversight. The speech links safety supervision with developing European AI capabilities.
The speech texts are marked as prepared versions. The sources disclose neither dispatch dates and individual replies nor established violations or fines. Additional requests and recommendations remain announced next steps.