Tag
security
2 dispatches
Five Eyes to Agentic AI: Assume It Will Misbehave
The Five Eyes cybersecurity agencies issued their first joint guidance on agentic AI, admitting their own evaluation frameworks can't fully assess the risks they're warning about.
Single-Prompt Safety Scores Are Measuring the Wrong Thing
Cisco tested 15 frontier AI models under multi-turn attacks and found safety bypass rates up to 88%, exposing a structural flaw in how the industry benchmarks model safety.