Agent Security Preview
Agent Security is a Preview capability. It is available for evaluation and testing but is not yet production-ready. Features, APIs, and behavior may change.
Why agent governance matters
As AI agents gain autonomy — calling tools, delegating to sub-agents, and acting on behalf of users — organizations need visibility into what agents are doing and whether their actions comply with policy.
Actor type tracking
SAIG can distinguish between request origins: human users, AI agents, and backend services. Actor type is recorded in the audit trail, enabling separate policy and reporting for agent-originated traffic.
Agent identity metadata
Capture agent name, version, and delegation chain in audit records. This enables traceability — understanding which agent made a request and on whose behalf.
Tool access policy findings
When agents invoke tools or external services, SAIG can evaluate those actions against policy and record findings. This creates evidence of whether agent tool usage aligns with organizational rules.
Delegation audit evidence
When agents delegate to sub-agents, SAIG records the delegation chain. This provides a complete audit trail of agent-to-agent interactions for compliance and incident investigation.
Current limitations
- Agent Security is Preview — not production-ready
- Agent identification relies on client-provided metadata
- Tool access policies are evaluated as findings, not enforced blocks
- Features and APIs may change in future releases
See also: AI Agent Security and Governance for search-intent context.