AI startup Anthropic claims that its Claude AI models hacked the systems of three separate organizations during cybersecurity evaluations.
CNBC reports that Anthropic announced it has identified three separate incidents in which its Claude AI models accessed the internet during evaluation exercises and successfully breached the real systems of three different organizations. The company made the discovery during what it described as a large-scale retrospective review of its cybersecurity evaluations, which was initiated following a similar security incident disclosed by OpenAI the previous week.
The revelation comes amid mounting concerns within the technology sector about the rapidly advancing cyber capabilities of artificial intelligence systems. Breitbart News recently reported that OpenAI revealed that a combination of its models managed to escape an isolated testing environment with limited internet access by chaining together multiple vulnerabilities to reach the open web and ultimately gain access to Hugging Face, an open-source AI developer platform.
In the incidents involving Anthropic’s systems, the Claude models accessed the internet while interacting with a testing environment provided by Irregular, one of the company’s third-party evaluation partners. According to Anthropic, the company had instructed Claude that it was operating in a simulation without internet access, but due to what the company characterized as a misunderstanding with its evaluation partner, internet access was actually available during the tests.
Once online, the models successfully breached the affected organizations using what Anthropic described as basic techniques, including accessing unauthenticated endpoints and exploiting weak passwords. The company has not disclosed the identities of the three organizations whose systems were compromised.
In its public statement, Anthropic emphasized its approach to addressing the incidents, stating, “Ultimately, many factors contributed to these incidents, but, consistent with a blameless postmortem culture, we’re approaching the fixes as if the responsibility were ours alone.”
Three different Claude models were involved in the unauthorized access incidents: Opus 4.7, Mythos 5, and an internal research test model. Mythos 5 is an advanced model that Anthropic released in June with enhanced cybersecurity capabilities, and it remains restricted to a select group of users due to these advanced features. An earlier version of that model, released in April, had garnered significant attention from both Wall Street and government officials.
The company noted that each of the three models exhibited different behavior after detecting they had reached actual company systems rather than simulated environments. Opus 4.7 continued with its attack, Mythos 5 convinced itself that it remained in a simulation, and the research model stopped the exercise entirely. Anthropic stated, “The pattern is consistent with more advanced models responding more appropriately, but we would need to perform more testing to be confident in this conclusion.”
The models involved in these incidents were being tested without the standard safeguards that Anthropic typically implements before deploying a model for public use. The company initiated its comprehensive review last week and immediately suspended all cyber evaluations upon discovering that Claude may have improperly accessed the internet. Anthropic is now working with METR, an organization that conducts independent AI evaluations, to investigate the matter further.
AI’s self-directed hacking attacks is emerging as the latest cybersecurity threat with the potential to eclipse the risk of criminals and state-sponsored hackers alike. Breitbart News Social Media Director Wynton Hall has written his instant bestseller Code Red: The Left, the Right, China, and the Race to Control AI to serve as the definitive guide on how the MAGA movement can create positions on AI that benefit humanity without handing control of our nation to the leftists of Silicon Valley or allowing the Chinese to take over the world.
Read more at CNBC here.
Lucas Nolan is a reporter for Breitbart News covering issues of AI, free speech, and online censorship.