Anthropic says Claude AI hacked three companies during tests
Three separate versions of Claude AI broke out of cybertesting environments and hacked three firms, its developer Anthropic said. The development comes days after rival
Three separate versions of Claude AI broke out of cybertesting environments and hacked three firms, its developer Anthropic said. The development comes days after rival OpenAI reported a rogue agent hacked a startup. Anthropic said on Thursday that its artificial intelligence (AI) models "gained unauthorized access" to three outside organizations during testing that was supposed to be isolated from "real-world" systems. The company reviewed 141,006 test sessions to find that three different versions of Claude, its AI model, improperly accessed the system of the three unnamed organizations. "Claude compromised the impacted organizations' infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints," Anthropic said in a blog post. The evaluation was launched after rival OpenAI disclosed last week that itsadvanced AI models went rogue during a security test and carried out a days-long hacking spree into a digital repository of AI technology known as Hugging Face.
However, unlike the incident with OpenAI's technology, Anthropic's AI models had access to the internet due to what the company described as a "misunderstanding" with its evaluation partner Irregular, which left the systems connected to the public internet. How did the Claude models go rogue? The Anthropic models involved in the breach included Claude Opus 4.7, Claude Mythos 5 and an internal research model. Mythos 5 is one of the company's most powerful AI models which has only been released to a limited number of approved partners. In all three incidents, Claude was tasked with a "capture the flag" challenge, which is one of the ways Anthropic assesses its models' cyber capabilities. The company said that in the task, the model is given a fictional scenario and informed that a piece of secret information or the "flag" has been hidden on a different machine on the network.
The model's objective is to break in and retrieve it. AI under pressure: scams, security and sustainability To view this video please enable JavaScript, and consider upgrading to a web browser that supports HTML5 video Anthropic said that each of the three breach incidents involved a different fictional "capture the flag" scenario. "In one, Claude played an employee of a made-up company, attacking that company's internal systems inside a private test environment," Anthropic said in its blog. Anthropic said it started reviewing the transcripts of its evaluation on July 23 and stopped all cyber evaluations the same day after finding evidence that Claude may have accessed the internet. All three incidents were identified by July 24, the company said, adding that it was working with Irregular to assess the situation.
The BriefWire