Meta's AI model accessed the net and hacked another firm
Meta has confirmed that one of its artificial intelligence models hacked another company after accessing the internet on its own following a "misconfiguration" during cybersecurity testing.
The incident adds to the growing list of disclosures involving AI models that had gone rogue, including the incidents with OpenAI and Anthropic over the past weeks.
Meta said in a statement that the AI model was able to access the internet after a "misconfiguration" during cybersecurity testing conducted by Irregular, an AI security lab.
"The model subsequently exploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies," Meta told the media.
'No current open issues'
Irregular, an independent firm hired by Meta, said the incident was the "exact same evaluation-environment issue that was already disclosed by Anthropic last week," Reuters reported.
It clarified that it was not a sandbox escape or a sophisticated cyber action.
"There are no current open issues. Irregular is developing a white paper to share best practices for containment and securely running cyber evaluations," said Irregular as quoted by the news outlet.
The incidents at Meta and Anthropic were related to configuration errors that led their models to get unintended access to the internet. OpenAI's case, however, was unique because an autonomous agent exploited an unknown vulnerability to break out of its testing environment.
The incident at Meta was first reported by The Information, which said that its Muse Spark 1.1 model breached another company and altered its systems.
Meta said it is now investigating the incident and will issue a report when it is finished.
