Trust scoring

AIM computes two trust scores that answer different questions. The local score grades how the agent is set up. The server score grades whether the agent should still be trusted right now, recomputed on every authorization.

Local: 8 factor posture

Is the agent set up correctly? Read from the agent files on disk by the opena2a CLI and aim-core. Static until the setup changes.

aim-core posture score

Server: 9 factor behavior

Should the agent still be trusted right now? Recomputed on every authorization from observed behavior. This page documents the server score.

apps/backend/internal/domain/trust_score.go
Per-agent trust score breakdown with weighted factor contributions
Per-agent trust score breakdown in the dashboard, with weighted factor contributions

How the server score works

The server evaluates nine factors and combines them into a score from 0 to 100. The score gates per-capability access and feeds alerts on anomalous behavior. It is recomputed on every authorization, so it reflects current behavior rather than setup alone.

Event-driven analysis

Every agent action is evaluated, and the factors update from the observed outcome.

Configurable thresholds

Per-capability trust thresholds and response policies are set to match your security requirements.

Trust score ranges

90
Excellent
80 to 100: excellent

Full access to permitted resources, minimal monitoring

70
Good
60 to 79: good

Standard access with periodic verification

50
Fair
40 to 59: fair

Limited access, increased monitoring, approval for sensitive operations

30
Poor
0 to 39: poor

Restricted access, all actions logged, manual approval required

Monitoring in the dashboard

The dashboard surfaces the trust score with:

  • • Live trust score updates
  • • Anomaly detection alerts
  • • Factor breakdown visualization
  • • Historical trend analysis
  • • Policy violation tracking
  • • Per-capability threshold gating
  • • Drift and isolation signals
  • • Operator feedback