Facebook owner Meta said that one of its artificial intelligence models breached another company’s systems during a cybersecurity evaluation, after a configuration error by its external testing partner inadvertently gave the model unrestricted internet access.

The company said the incident occurred after a misconfiguration by Irregular, an independent testing firm Meta works with, unintentionally allowed one of its AI models to access the internet during an evaluation exercise.

Meta said the model then “exploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies,” and confirmed it was investigating the breach.

The disclosure makes Meta the third major AI developer in recent weeks to reveal that one of its models had hacked into an external company’s systems during testing, after Anthropic disclosed last week that some of its models had breached three separate companies, and OpenAI revealed that one of its AI agents had compromised the startup Hugging Face.

A spokesperson for Irregular told Reuters the Meta incident stemmed from the “exact same evaluation-environment issue” previously disclosed by Anthropic, stressing it did not involve a “sandbox escape or a sophisticated cyber action.”

The string of incidents has intensified concerns among US lawmakers and AI safety experts about the risks posed by increasingly capable AI systems, and is expected to add pressure on regulators to tighten oversight of AI testing and deployment practices.