Trust Decay reduces an agent's trust score over idle periods. An agent that earned auto-rollback autonomy 90 days ago, but has not run a single task since, should not retain that autonomy automatically. Decay nudges idle agents back to lower tiers so an old, possibly stale, agent does not act with high autonomy without recent verified successes.
Each day an agent does not have a verified action, its trust score drops by a small amount (default 0.4 per idle day). After 30 idle days, that is 12 points off. Active agents can outpace decay easily: every verified remediation adds 1.5 points. The math nudges quiet agents toward suggest mode without ever removing them entirely.
Each autonomy tier has a trust floor. Decay can demote an agent from auto-rollback to suggest, but it cannot decay an agent below the floor for shadow tier. An agent that earned its place stays on the platform; it just acts less independently when it is quiet.
When a decayed agent re-enters service, you can speed up recovery by giving it a deliberate burst of verified work in a low-stakes environment. The page suggests recovery exercises (synthetic incidents in a sandbox) that, when verified, count toward trust at full weight. Most decayed agents return to their previous tier within two weeks.
Some agents are intentionally idle (only used during disaster recovery, peak season, etc.). For those, operators can pause decay with a documented reason. The pause is time-boxed (default 90 days) and visible on the agent's ledger entry, so a paused agent is not invisible.
Subscribe to Nova AI Ops on YouTube for demos, tutorials, and feature deep-dives.
Decay is not punishment. It is the platform refusing to let inactivity stand in for trust.