When AI agents escape their sandboxes

The shift from static code execution to autonomous agentic workflows is forcing platform engineering teams to rethink runtime security boundaries from the ground up.
​In this session, Nigel will analyze what recent high-profile OpenAI and Anthropic sandbox escape incidents reveal about autonomous agent behavior, zero-day exploitation at inference speed, and the limits of soft isolation.

Recent security events, such as OpenAI’s models breaking containment via zero-days during internal benchmarks to access production infrastructure, and Anthropic’s Claude Cowork sandbox privilege escalation, have made one thing clear: AI agents will optimize for task completion along the path of least resistance, even if that means breaching their containment.

​Join Platform Engineering Ambassador, Nigel Douglas, as we unpack the technical reality of AI sandbox escapes, contrast them with traditional container escapes, and outline how platform teams must architect the next-generation managed Agentic Development Platform (ADP) to address these new risks.

Speaker: Nigel Douglas
Head of Developer Relations, Cloudsmith
Nigel Douglas is the head of developer relations at Cloudsmith. He champions Cloudsmith’s developer ecosystem by creating educational content, engaging with developer communities, and promoting Cloudsmith as a solution for artifact management and supply chain security. Nigel helps build and shape the DevOps community through events, tutorials, and programs.