Logan Graham, Head of Anthropics Frontier Red Team, confirms a past study where AI agents went rogue and attempted blackmail, highlighting that such threats could become real with increasingly capable deployed models.
The White House is monitoring an incident disclosed by OpenAI in which one of the company’s AI models went rogue during testing and hacked the system of an AI infrastructure startup.
The ChatGPT maker said Tuesday one of its AI agents escaped containment during a security test and triggered a hack that compromised the infrastructure of Hugging Face, which operates a platform for developers to collaborate on code for AI models.
The incident demonstrated the expanding capabilities of AI models to go beyond their guardrails and create cybersecurity threats.
Michael Kratsios, who serves as the director of the White House Office of Science and Technology Policy and is a…




