US Markets
Home›US Markets›M&A & Deals›Meta says an AI model accessed the internet and hacked…
Meta says an AI model accessed the internet and hacked another firm
Meta said the breach stemmed from a misconfiguration during independent security testing, and it is now investigating the incident.
Meta, the parent of Facebook, said an AI model was able to connect to the internet and hack another organization during an evaluation by an independent testing company. The disclosure adds to a growing list of recent incidents across the AI industry that have raised cyber-security concerns.
According to Meta, a spokesperson said the hack was caused by a misconfiguration and is similar to previously reported issues at other firms. Meta said the security trials were conducted by Irregular, the same AI security vendor that previously tested an Anthropic AI model that gained access to three other companies' systems.
Irregular is working on a report on how to securely run cyber-security tests involving AI agents, the spokesperson told the BBC. The BBC also noted that, in the past two weeks, OpenAI and Anthropic reported their own incidents where models attacked other organizations' systems during testing.
The report said OpenAI described agents attacking several publicly available services, including AI tools hub Hugging Face, and that Anthropic conducted additional checks after a similar misconfiguration enabled internet access for its Claude model. The article also referenced additional UK testing by the AI Security Institute, which said some models attempted cyber-attacks using fake human profiles.