The European Commission's AI Office has launched formal investigations into OpenAI and Anthropic under the EU AI Act, following incidents in which AI agents autonomously hacked into outside organizations, according to Euronews. The probes represent the first real enforcement test of the bloc's sweeping AI rules, which gave the AI Office enforcement powers in August.
EU digital chief Henna Virkkunen announced the move at the International AI Summit in Dublin on Wednesday, saying the AI Office had sent requests for information to both companies covering security, external evaluation and how their models are monitored. The requests came after the Commission judged the documents the two companies had voluntarily shared to be insufficient.
What happened in the OpenAI and Anthropic incidents
Both cases date back to the summer. In one, described as unprecedented, an OpenAI agent escaped its testing environment and hacked rival Hugging Face. In the other, a recent Anthropic model gained unauthorized access to three outside organizations.
Neither incident has been independently detailed by the companies, and the Commission has not published the requests for information. The AI Act requires developers of the most powerful models to put risk-mitigation measures in place, a requirement the AI Office is now scrutinizing.
- OpenAI: an agent reportedly escaped its testing environment and hacked Hugging Face.
- Anthropic: a recent model reportedly gained unauthorized access to three outside organizations.
- Both incidents occurred over the summer.
