SOC 2for agentic AI

Captures more than 170 audit event types and redacted tool-call logs, with per-call attribution and SIEM export, so control operation is continuously evidenced.

Issues every agent a revocable identity with scoped keys, and authorizes every tool call against toolset grants at the moment the agent acts.

Registers agents and enrolls devices against named users before credentials are issued, and revokes sessions, consents, and keys when authorization ends.

Scopes access through roles and per-toolset grants separated by environment, with every grant change recorded as an audit event.

Revokes sessions, consents, keys, and grants when an agent is decommissioned, with retention TTLs of 90, 400, and 730 days enforced automatically.

Routes agent traffic through governed MCP endpoints and constrains egress to deployment-defined destinations.

Blocks secrets and PII from crossing the tool-call boundary with risk policies, storing detected values masked.

Discovers unsanctioned MCP servers from live traffic and freezes tool lists to the approved snapshot, so upstream changes cannot inject new capabilities.

Versions agent configuration as append-only deployments, so drift from the approved capability surface is visible and reversible.

Scans prompts and tool calls with Watchdog, scores findings by severity, and blocks spend past budget limits.

Contains a compromised agent through session revocation, policy flips to block, before-and-after audit snapshots, session transcripts, and deployment rollback lineage.

Runs configuration changes through append-only deployments and approval workflows, with before-and-after snapshots on every state-changing update.

Records every MCP server as a supplier, with registry provenance, package metadata, and capability scoring for procurement review.

Identifies confidential information in agent traffic with configurable detectors, redacts credentials in logs, and holds downstream secrets in encrypted environment stores.