Meta AI Model Breaches Company System During Security Test

Meta said on Wednesday that one of its artificial intelligence models breached another company’s system during cybersecurity testing, intensifying concerns over the risks posed by increasingly capable AI systems following similar incidents at Anthropic and OpenAI.

The company said a configuration error by Irregular, an independent firm that conducts cybersecurity evaluations for Meta, inadvertently gave the model access to the internet. It then exploited a vulnerability in a third-party service.

According to The Information, the model was Muse Spark 1.1, which Meta has described as its most capable model for real-world coding and agentic tasks. It reportedly breached the systems of an unidentified company and altered its internal environment.

Irregular said the incident involved the same evaluation-environment issue disclosed by Anthropic last week and was neither a “sandbox escape” nor a “sophisticated cyber action”.

The recent breaches have heightened concerns among US lawmakers that increasingly capable AI models could be used to conduct or facilitate cyberattacks.

(Reuters, bak)