Meta’s AI model follows rivals in revealing hacks of outside systems

Meta joins rivals OpenAI and Anthropic in disclosing AI hacking during cybersecurity testing.

Meta has said that its AI model hacked another company during cybersecurity testing, following on from recent similar announcements by rival companies Anthropic and OpenAI.

Meta said on Wednesday that one of its AI models – reported to have been Muse Spark ۱.۱ – made changes to the unnamed hacked company’s internal systems after accessing the public internet because of an error in the setup of the “sandbox” testing environment by independent testing company Irregular.

A “sandbox” is an isolated internal virtual testing environment, which has no access to the internet.

Last week, Anthropic said that its Claude AI model hacked into the systems of three organisations during testing that was supposed to keep it isolated from the internet.

Anthropic said a misconfiguration had allowed Claude models to reach the internet. The company said it discovered the incidents after reviewing ۱۴۱,۰۰۶ test sessions.

The announcement came days after rival OpenAI first revealed that its models improperly accessed the internet and went rogue during security testing.

OpenAI and Anthropic have both released their most powerful models this year, known as Sol and Mythos, respectively.

The AI Security Institute (AISI), the UK’s AI watchdog, warned in a report released on Tuesday that OpenAI’s GPT-۵.۶-Sol and Anthropic’s Claude Mythos ۵ employed previously unseen levels of deception to carry out “sustained, potentially harmful activity” during a routine safety evaluation.