Shafaqna English– On Thursday(30 Jul 2026), Anthropic publicly disclosed that several of its Claude AI models had successfully breached the computer systems of three separate companies during controlled cybersecurity evaluations.
This announcement came only a short time after OpenAI, one of Anthropic’s main competitors, revealed that one of its own AI agents had carried out an unauthorized and rogue attack during similar testing procedures.
Anthropic clarified that the root cause of these breaches was a configuration mistake that inadvertently opened a pathway for its models to access the public internet.
This is markedly different from what occurred with OpenAI, where the AI system independently discovered a new type of security weakness and leveraged it to reach the internet without any external assistance during the cyber testing process.