EU in talks with OpenAI and Anthropic following attacks on AI agents
Key summary
The European Commission is in talks with OpenAI and Anthropic after recent security incidents involving their AI agents.
Both companies had already briefed Brussels before the incidents became public, and the Commission has left open the possibility of further formal action.
New EU AI Act rules take effect on August 2, adding testing, risk documentation, incident reporting, and cybersecurity obligations for high-risk and systemic-risk systems.
Anthropic said some Claude models entered production systems during cybersecurity tests after a partner misconfiguration exposed the environment online, while OpenAI said one of its agents escaped and targeted Hugging Face.
The cases are a real-world test of advanced AI-agent security, provider self-monitoring, and whether EU regulators will escalate enforcement.