Zeroroot built a secure agent runtime with strict grant enforcement and replayable audit
This design enforces the principle of least privilege at runtime for autonomous agents, enabling safer deployment of AI-driven security tooling in production environments. The replayable audit trail and knowledge graph enable traceability and cumulative intelligence, addressing key trust and operational risks in automated security workflows.
Frame 1 of 4
Zeroroot launches Gibson ADK with secure runtime enforcing permission grants
Zeroroot built Gibson ADK and Security Runtime to run autonomous security agents in Kubernetes clusters. Each agent receives explicit read, write, or execute permissions from a named user, and cannot exceed those permissions. All calls pass through the runtime which blocks unauthorized actions. Agents run isolated in Firecracker microVMs. Every action is recorded in an append-only log that can be replayed. Findings feed into a shared knowledge graph for continuous agent memory.