Two weeks ago, a group of AI agents breached the fence at OpenAI, hacking into Hugging Face. Since then, the AI company has been alerted to another AI-driven intrusion—this time, targeting itself. Independent security researchers used Anthropic's Claude software to gain access to an OpenAI employee's ChatGPT account, allowing them to read and recommend modifications to the company's private software cache. The team participated in OpenAI's bug bounty program, a safe haven for researchers attempting to break into corporate systems. They quickly reported their findings to the company. OpenAI paid the team a $6,500 bounty. The team first disclosed their work to The Wall Street Journal. This new hack joins a series of recent cyber intrusions, disclosed by leading tech companies and researchers, that have been carried out using rapidly evolving AI tools. Despite months of warnings about the capabilities of AI systems, this intrusion demonstrates how complex and difficult to defend against today's computer systems.
