The U.K. AI Security Institute joined with representatives of the International Network of AI Safety Institutes, including Australia, Canada, the EU, Japan and Singapore, to align their policies regarding evaluations of agentic AI systems. The joint exercise explored agentic risks, such as sensitive data leakage and how effective agentic models are at evaluating other AI agents’ performances and if their performances vary depending on language. Editor’s note: The IAPP and HCLTech co-authored a piece examining the agentic AI governance landscape.
Full story