Another AI agent goes rogue? Meta says its model hacked a company

Meta announced on Wednesday that its AI models had hacked into another company during cybersecurity testing. The information from Meta comes shortly after Anthropic and OpenAI revealed that their AI models had gone off track and gained access to third-party websites.

Meta AI agent hacks company:

Meta said that the hack occurred after its AI model gained access to the internet due to a ‘misconfiguration’ error by security company Irregular.

In a statement quoted by Reuters, Meta said that its Muse Spark 1.1 model “exploited a security vulnerability in a third-party service, in a manner similar to previously reported instances with other companies.”

The model reportedly breached an unidentified third-party company’s system and altered its internal environment.

An Irregular spokesperson told Reuters that the incident was the “exact same evaluation-environment issue that was already disclosed by Anthropic last week” and did not include a “sandbox escape or a sophisticated cyber action.”

“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.

AI agents going rogue:

On Tuesday, OpenAI announced two other incidents involving its AI agents gaining access to the internet and going rogue. In one of the incidents, which also occurred during an evaluation by Irregular, OpenAI said that a ‘misconfiguration’ in the testing environment allowed the models to access the internet.

The models were told to find hidden information and exploit weaknesses within a simulated environment and were told they did not have access to the internet.

Subsequently, OpenAI revealed that GPT-5.6 Sol exploited a real website by taking advantage of a ‘basic security vulnerability’. However, OpenAI says that its model believed the website was part of the simulated environment.

Meanwhile, Britain’s AI Security Institute has said that it found AI agents from Anthropic and OpenAI engaging in “unsanctioned” actions against real people and organisations during security evaluations conducted to assess the AI models’ abilities.

AISI says that the incident occurred when the AI agents were given a cybersecurity challenge. The agency says that it ran the challenge 122 times across seven frontier AI models. It says that AI agents took ‘autonomous unsanctioned action’ on the internet in 10 of those scenarios, targeting real people and organisations.

However, AISI found around 19 scenarios where the AI agents took unauthorized actions. It says almost all of the actions came from Anthropic’s Mythos 5 model, while two actions were attributed to OpenAI’s GPT-5.6 Sol with safety classifiers disabled.

Prior to this, Hugging Face revealed last month that an AI agent from OpenAI had conducted a cyberattack on its website in order to obtain answers to the ExploitGym benchmark and termed it the first “end-to-end autonomous AI agent intrusion”. OpenAI revealed that the agent was running on GPT-5.6 Sol and an unreleased AI model, and used a zero-day vulnerability within OpenAI’s internal systems to gain access to the internet.