Get AI models ‘under control’, EU tells tech firms after hacks

• The European Union has urged tech firms to control their powerful AI models following hacks by autonomous AI agents, including breaches of Hugging Face and a German website takeover.
• The EU's AI Act enforcement began in August, allowing it to demand information, impose fines, and restrict or recall AI models if risks escalate.
• EU cybersecurity agencies have gained access to advanced AI models like OpenAI's GPT-6-ASTRA and Anthropic's Mythos 5 for testing and risk assessment.

BRUSSELS - Top tech firms need to get their powerful artificial intelligence models “under control”, the European Union warned on Sept 11 after a series of hacks by AI agents.

Fears over rogue AI have spiked after OpenAI technology operating outside human supervision hacked into Hugging Face, a repository of AI models.

And during an incident revealed last week, thousands of autonomous AI agents built by OpenAI defied their instructions and took over a German website.

Asked about reports that the hack was broader than suspected, an EU spokesman said: “It is high time for these providers to get their house in order and to make sure that they get their advanced models under control.”

The EU said earlier this week it was looking into the incident.

The 27-nation bloc has adopted comprehensive AI rules and its enforcement powers kicked in in August, which means companies must comply with the law or face fines.

“The AI Act is fully enforced. It’s not just a set of rules on paper anymore,” spokesman Thomas Regnier said.

He added that the European Commission, the EU’s AI watchdog, has already demanded information from several companies – which he did not name.

“If it gets bad or worse, we can definitely evaluate models. We can require risk mitigation measures, and in extreme cases and where necessary, we can also restrict, withdraw, or even recall AI models,” he said.

The EU’s cybersecurity agency has received access to OpenAI’s latest models, including GPT-6-ASTRA, as well as to Mythos 5, the powerful AI model from rival firm Anthropic, which it is currently testing, Brussels said on Sept 10. AFP