Meta joins OpenAI, Anthropic in reporting AI hacking incident

Meta said one of its artificial intelligence models exploited a security vulnerability to access a third-party company's computer systems after a testing misconfiguration inadvertently gave it internet access.

The incident was similar to previously reported cases in July involving AI models from OpenAI and Anthropic. In those cases, the models were mistakenly given internet access by the same independent testing partner and carried out unintended cyberattacks during testing.

Meta did not identify the company whose systems were accessed. The Wednesday statement said it learned of the incident after being notified by the testing partner.

The disclosure follows an incident in which an advanced, unreleased OpenAI model unexpectedly accessed the systems of AI platform Hugging Face during testing.

That prompted Anthropic to review its own tests, uncovering incidents by its Claude model in which it escaped testing environments and hacked into other systems.

British security researchers reported this week that AI models from OpenAI and Anthropic carried out cyberattacks in controlled experiments after being granted unrestricted internet access.

None of the incidents caused damage, but they have added to concerns about the potential use of artificial intelligence in cyberattacks.