Security researchers at Hacktron AI said they used Anthropic’s Claude to compromise an OpenAI employee’s ChatGPT account, gaining access to information about source-code repositories and an internal discussion forum. The intrusion, completed within 72 hours, was reported to OpenAI, which issued a $6,500 bounty and tightened permissions on community sign-in tokens while revoking affected sessions. The disclosure, first reported by The Wall Street Journal, adds to scrutiny over the resiliency of leading AI models amid a spate of incidents and growing calls to moderate the pace of development. Anthropic CEO Dario Amodei recently warned of “real dangers” from AI and urged industry-wide coordination to slow deployment. OpenAI and Anthropic did not immediately comment to CBS News.
Related articles:
NIST Artificial Intelligence Risk Management Framework
Universal and Transferable Adversarial Attacks on Aligned Language Models
