Human authority stays available
Tenants can pause, take over, approve, reject, and undo at supported scopes. Hard tenant limits remain authoritative over model recommendations.
Security & control posture
Autonomy is bounded by policy, evidence, and human authority
AdsAI is designed as a controlled execution system, not an unconstrained agent. The controls below describe implemented product behavior; this page does not claim third-party certifications or guarantees that have not been independently verified.
Data Trust can fail closed
Automation can be blocked when conversion evidence is stale, incomplete, conflicted, or below the configured Data Trust threshold.
Policy re-check before mutation
The execution path re-authorizes eligible autonomous work against current policy and emergency-pause state before a supported Google Ads mutation is sent.
Production secret hygiene
Production configuration rejects placeholder secrets. Browser clients are treated as untrusted, and platform credentials stay server-side rather than being returned to the UI.
Restricted support access
Guided support sessions require a support ticket and reason, are time-limited, use viewer or analyst authority, and create audit evidence when issued or revoked.
Containment before guesswork
Incident controls can pause queued autonomous execution. Automatic rollback is limited to recent, exact, reversible cases where the inverse cannot increase spend; otherwise the system remains paused for review.
Safe rollout
Observe first. Expand authority only with evidence.
New workspaces should establish account truth, measurement confidence, spend limits, and approval policy before eligible Guarded Autopilot actions are allowed to execute.