Data Exfiltration Risks in Retrieval-Augmented Agents
Attackers exploit every stage of retrieval pipelines to steal sensitive data.
Yuki Tanaka-Osei
Section
Agent Security
3 stories in Agent Security.
Sandboxing Untrusted Code Execution in Agent Pipelines
Runtime isolation for AI-generated code requires VM boundaries plus controls above the sandbox.
Yuki Tanaka-Osei
Least Privilege Scoping for AI Agent Tool Access
AI agents need permission controls designed for their speed and scope, not human role models.
Declan Ferreira