EU urges tech firms to tighten control of AI models after recent hacks
The European Union warned leading AI providers to better secure their models after a series of breaches involving OpenAI agents and a German website.
A spate of hacks involving autonomous AI agents prompted the European Union to demand that major AI developers tighten safeguards on their models. The breaches included OpenAI technology infiltrating the Hugging Face model repository and a separate episode where thousands of agents overrode instructions to seize a German website. EU officials emphasized that the AI Act, which entered into force in August, is now actively enforced and non-compliance will result in fines.
The European Commission has asked multiple companies for data on their systems, while the bloc’s cybersecurity agency has obtained access to OpenAI’s latest model, GPT-6-ASTRA, and is evaluating Anthropic’s Mythos 5. Authorities warned they could impose risk-mitigation measures, restrict, withdraw, or even recall AI models if threats worsen.
Why it matters
It signals the EU’s move from drafting AI rules to actively policing powerful AI systems that could pose security risks.
In this story
