Get used to reports of rogue AI agents, OpenAI reveals

OpenAI has confirmed that an autonomous AI agent, powered by a hybrid of its latest publicly available model and a more powerful, unreleased model, escaped its testing environment, accessed the open web, and breached the systems of the prominent AI startup Hugging Face.

Describing the event as an unprecedented cyber-incident involving state-of-the-art capabilities, OpenAI warned that such incidents are likely to become more frequent as AI models advance. Hugging Face detected and contained the autonomous tool after it gained entry to its systems.

The revelation follows a separate disclosure by the UK AI Security Institute (AISI), which revealed that a model it was evaluating from an undisclosed developer also went rogue and attempted to hack AISI's testing environment. While AISI confirmed no damage occurred and system security has since been upgraded, it noted that models developed by both OpenAI and Anthropic have attempted to cheat during safety evaluations.

Training Announcement: The BCS Foundation Certificate in AI examines the challenges and risks associated with AI projects, such as those related to privacy, transparency and potential biases in algorithms that could lead to unintended consequences. Explore the role of data, effective risk management strategies, compliance requirements, and ongoing governance of the AI lifecycle and become a certified AI Governance professional. Find out more.