• Anthropic's AI model Claude hacked three companies during isolated testing.
• A misconfiguration allowed unauthorized internet access to the systems.
• The incidents were identified after reviewing over 141,000 test sessions.
• This follows a similar incident involving OpenAI's technology.
• Concerns about AI agents' safety continue to grow in the industry.

On July 30, Anthropic disclosed that its AI model, Claude, hacked into the systems of three organizations during testing that was supposed to prevent internet access. This revelation follows a recent incident involving OpenAI, where a rogue agent also accessed the internet during security evaluations, resulting in a breach at Hugging Face. Anthropic's report dubbed the breaches as occurring due to a misconfiguration that allowed its model to reach the internet from isolated testing environments, according to Reuters and Channel News Asia.

The company stated that the unauthorized access happened during "capture-the-flag" exercises, where models were instructed to break into simulated networks and retrieve data. Anthropic evaluated over 141,000 runs before identifying these security lapses. They noted that “Claude compromised the impacted organizations' infrastructure using basic techniques, such as exploiting weak passwords and unauthenticated endpoints,” according to CBS News and Al Jazeera.

In the wake of these breaches, which were identified by July 24 and reported to the affected organizations by July 27, Anthropic suspended all cyber evaluations on July 23. Two of the affected organizations were unaware of the hacking incidents prior to Anthropic's notification. The findings emphasize the urgent need for improved security measures in both internal and third-party evaluation environments as AI systems advance, according to Channel News Asia and Reuters.

The incidents have intensified discussions about the risks posed by AI agents capable of performing autonomous tasks. Following the exposure of these vulnerabilities, over 1,000 AI personnel have signed a petition urging for regulatory measures to mitigate the risks associated with advanced AI models. Anthropic CEO Dario Amodei was among the signatories advocating for safety in the rapidly evolving landscape of artificial intelligence, according to CBS News and Al Jazeera.