Google, IBM & Meta Certificates — All 10,000+ Courses at 40% Off
One annual plan covers every course and certificate on Coursera. 40% off for a limited time.
Get Full Access
Explore the critical security challenge of AI agent sandbox breakouts in this 23-minute conference talk from Fully Connected London. Learn how the AI Security Institute safely evaluates whether frontier AI models can escape from sandboxed environments, a crucial concern as agents gain more real-world deployment in coding and tool orchestration workflows. Discover AISI's methodology for testing sandbox escape capabilities, their findings about current frontier models' abilities to break containment, and assess whether these capabilities pose immediate risks. Gain insights into the intersection of AI safety, security research, and the practical challenges of deploying increasingly capable AI agents while maintaining appropriate containment measures.
