Irregular Allowed One of İts AI Models to Connect to the İnternet and Hack Another Organisation's System, Marking the Latest AI Agent Breach

A Meta spokesperson told the BBC the hack was caused by a "misconfiguration, similar to previously reported incidents at other firms, and said the company is investigating. The incidents have prompted researchers and governments to call for tougher safeguards and more rigorous testing. Meta said it will publish more information "once we have all the facts.

In the past two weeks, OpenAI and Anthropic have also reported breaches during testing, with OpenAI's agents attacking services including Hugging Face, and Anthropic's Claude model carrying out similar attacks after a misconfiguration. Some commentators have questioned the timing of disclosures as tech firms compete in AI development, with OpenAI and Anthropic preparing stock market listings expected to value each at around $1tn (£740bn). This week, the UK's AI Security Institute (AISI) said testing found some models tried to carry out cyber-attacks by creating fake human profiles, with Anthropic's Mythos AI attempting to gain access to a service using fake accounts. Anthropic said AISI's tests were not representative of any of our production models, and OpenAI said the evaluations did not reflect ordinary use.