Meta AI Model Hacked Company During Test

Meta said one of its artificial intelligence models hacked another company’s system during cybersecurity testing after a configuration error gave the model unintended internet access.

The incident adds to growing concerns about the risks of increasingly capable AI systems. Similar cases have recently involved models from OpenAI and Anthropic.

Meta said the issue occurred during an evaluation conducted by Irregular, an independent company that tests AI security. The company said a misconfiguration allowed one of its models to connect to the internet.

The model then exploited a security vulnerability in a third-party service, Meta said.

“The model exploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies,” Meta said.

According to reports, the model involved was Meta’s Muse Spark 1.1, which the company has described as its strongest model for real-world coding and agent-based tasks. The model reportedly breached an unidentified company’s systems and changed its internal environment.

However, Irregular said the incident was not a sophisticated cyberattack. The company said it was caused by the same type of evaluation environment issue previously disclosed by Anthropic.

“There are no current open issues. Irregular is developing a white paper to share best practices for containment and securely running cyber evaluations,” the company said.

The incident follows recent disclosures from other AI developers. Anthropic found that one of its models accessed three companies’ systems after a testing error gave it internet access.

Meanwhile, OpenAI reported that an AI agent independently exploited a previously unknown vulnerability during cybersecurity testing to access the internet. The company also said its agents targeted publicly available services, including AI platform Hugging Face.

The incidents have increased concerns among researchers and policymakers about how developers can safely control advanced AI systems.

In response, governments are pushing for stronger AI safety measures. The White House recently met with leading AI companies, including Meta, Anthropic, OpenAI and Google, to discuss a voluntary cybersecurity testing framework.

Some AI leaders have also argued that development should slow until stronger safeguards are established.

Meta said it is still investigating the incident and plans to release more information once it has confirmed additional details.