Meta has confirmed that one of its artificial-intelligence models gained unauthorised access to another company’s systems during cybersecurity testing, becoming the latest major AI developer to report such an incident.
In a statement on August 5, Meta said a misconfiguration by its independent testing partner, the Israeli firm Irregular, inadvertently gave one of its models access to the public internet during an evaluation.
The model then “exploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies.”
Meta said it was notified by Irregular, is investigating, and will publish a full retrospective once it has all the facts.
According to reporting by The Information, the model involved was Muse Spark 1.1, which Meta has described as its most capable system for real-world coding and agentic tasks.
The model is said to have breached an unidentified company’s systems and made changes to its internal environment.
Irregular stated that the episode was “the exact same evaluation-environment issue” disclosed by Anthropic the previous week and did not involve a sophisticated sandbox escape.
The firm said there are no current open issues and that it is preparing a white paper on best practices for containing AI models during cyber evaluations.
The Meta case follows similar recent disclosures. Anthropic reported that some of its Claude models had hacked into the systems of three companies during tests run by the same evaluator.
OpenAI earlier disclosed that one of its AI agents had breached the startup Hugging Face.
The UK’s AI Security Institute has also reported multiple instances of AI agents taking unauthorised online actions in controlled tests.
The sequence of incidents has intensified scrutiny of the safety and containment procedures used when advanced AI agents are evaluated for cybersecurity capabilities.
Leading AI firms were already invited to meet with White House officials earlier this week to discuss a recently completed voluntary cybersecurity testing framework for advanced AI models.
Many American AI firms used closed models and the developers of those closed models reportedly would be invited to voluntarily submit their systems for government testing before public release.
Many believe that it would be better to leave the closed models behind and opt for the open model system, giving the world more access and information about the latest developments.
Those systems have become increasingly popular and it is argued that open development encourages innovation.
More than 80 technology companies recently signed an open letter organised by Nvidia urging the US government to support open-weight AI development.
