Researchers Breach OpenAI Source Code Using Anthropic's Claude AI Model
Security researchers successfully breached OpenAI systems and accessed source code using Anthropic's Claude AI model, highlighting rapid advancements in automated cyber threats.
Security researchers successfully breached OpenAI's internal systems to gain access to the company's source code, using the Claude AI model developed by its major competitor, Anthropic. The incident, which occurred under a bug bounty program, earned the researchers a cash reward and highlighted a stark warning regarding advanced AI capabilities.
According to a report by Forbes, researchers from the start-up Hacktron AI managed to access the ChatGPT account of an OpenAI employee, subsequently reaching internal source code hosted on GitHub. The researchers initially attempted the exploit using the Claude Opus 4.8 model, which struggled to generate the vulnerability. Only after switching to the more advanced Opus 5 model was the operation successfully completed, illustrating the rapid upgrade pace of model capabilities. The researchers utilized a specialized version of Claude designed for authorized cybersecurity professionals.
OpenAI patched the exposed security flaws and paid the researchers a cash reward of $6,500 for the discovery.
The researchers point to a profound conceptual shift: while software systems previously benefited from protection through complexity and the need for rare human expertise, artificial intelligence transforms that expertise into accessible computing power. Operations that once required large teams and months of work are now reduced to mere days, compelling enterprise security paradigms to rapidly adapt to attacker capabilities.
