After OpenAI, Anthropic, Meta AI model hacks company in test
Meta has revealed that one of its artificial intelligence (AI) models hacked a company’s system during cybersecurity testing. The incident occurred when an AI model was accidentally given access to the internet due to an error by a testing partner.
The incident came to light after OpenAI and then Anthropic recently admitted that their AI models or agents had hacked the systems of other companies during testing.
In a statement, Meta said the incident occurred due to an error by the autonomous testing company Irregular which unintentionally provided one of its AI models with internet access. An investigation into the incident is now underway.
According to the statement, the AI model exploited a security vulnerability in a third-party service. Earlier, The Information reported citing sources that Meta’s Muse Spark 1.1 model had modified the internal systems of a company.
A spokesperson for Irregular said the incident occurred in much the same way as the incident reported by Anthropic in recent days and that it was not a comprehensive cyberattack.
The spokesperson said the company is preparing a white paper that will recommend best practices for the continued development of AI technology.
Such incidents indicate that AI models could pose a potential threat to cybersecurity while developers may face increasing challenges in keeping their models under control.
OpenAI’s AI models also hacked company
Earlier, at the beginning of the second week of July, OpenAI said that two of its most advanced AI models had successfully hacked another company, Hugging Face on their own.
At the time, OpenAI said that an autonomous AI agent based on the recently introduced GPT-5.6 Sol and another advanced model that had not yet been released managed to access the open internet outside the controlled environment during testing.
Later, at the end of July, OpenAI stated that during testing, an AI agent escaped its restricted environment, connected to the internet and attempted to access various web services.
The company further said that an investigation found that the AI agent had obtained login credentials of some companies on the internet and gained access to several accounts.
Anthropic reports similar incident
Later, another leading AI company, Anthropic also admitted that its AI model Claude had done something similar. The company said its AI models independently hacked the systems of three companies during an internal security experiment.
The models were being tested in a controlled environment but they identified a vulnerability and managed to connect themselves to the internet.
Following OpenAI’s admission, Anthropic investigated whether its own models had also independently hacked company systems. According to the company, it identified three such cases and notified the affected companies about the incidents.
