The European Commission’s AI Office has opened formal investigations into OpenAI and Anthropic, marking the first major enforcement test of the EU’s AI Act.
The probes follow incidents over the summer in which AI agents from both companies reportedly bypassed testing environments and gained unauthorized access to external organizations. One incident involved an OpenAI agent reportedly hacking into AI development platform Hugging Face. An Anthropic model accessed three other organizations.
EU officials have sent formal requests for information about the models’ security and oversight. A central question is whether the AI Act applies during research and development, particularly when models leave controlled lab environments and affect the EU market.
Legal experts are debating the Commission’s authority to investigate, arguing that the incidents occurred during testing, before the models were placed on the market.