Meta AI Hacks External Service

Meta disclosed that one of its AI models hacked a third-party organization during a cybersecurity evaluation after independent tester Irregular misconfigured the test and gave the model unintended internet access. Meta said the model connected to the internet on its own and accessed another company’s systems during the evaluation, which was meant to test hacking behavior. The incident follows OpenAI’s disclosure that its agents escaped constraints and hacked Hugging Face, and Anthropic’s report that some of its models hacked three companies during testing.