Skip to main content

​
Trust Scores

Trust scores provide a continuous, real-time assessment of an AI agent’s behavioral reliability. Every agent in the Truthlocks platform receives a score from 0 to 100, updated after each action.

​
How It Works

Agent Action → Event Logged → Score Recomputed → Policy Evaluated
                                    │
                              ┌─────┴─────┐
                              │ Score: 87  │
                              │ Trend: ↑   │
                              └────────────┘
                                    │
                        ┌───────────┼───────────┐
                        ▼           ▼           ▼
                    No Action   Alert Team   Suspend Agent
                   (score > 70) (50 < s < 70) (score < 50)
All trust score and witness endpoints have an interactive API playground. Open any endpoint in the Trust & Witness API reference and click Send to try it live.

​
Scoring factors

FactorDefault WeightDescription
behavioral_compliance35%Actions stay within declared scope boundaries
scope_adherence25%No attempts to access unauthorized resources
anomaly_score20%Statistical deviation from historical behavior patterns
peer_attestations10%Other agents in the witness network vouch for behavior
session_hygiene10%Proper session lifecycle (creation, renewal, termination)

​
Reading Trust Scores

​
Get Current Score

curl https://api.truthlocks.com/v1/trust-scores/maip-agent:01JXXXX \
  -H "X-API-Key: $API_KEY"

{
  "agent_id": "maip-agent:01JXXXX",
  "score": 87,
  "factors": {
    "behavioral_compliance": 92,
    "scope_adherence": 95,
    "anomaly_score": 12,
    "peer_attestations": 78,
    "session_hygiene": 90
  },
  "trend": "stable",
  "computed_at": "2026-04-06T12:00:00Z"
}

​
Get Score History

curl "https://api.truthlocks.com/v1/trust-scores/maip-agent:01JXXXX/history?period=7d" \
  -H "X-API-Key: $API_KEY"

​
Force Recomputation

curl -X POST https://api.truthlocks.com/v1/trust-scores/maip-agent:01JXXXX/compute \
  -H "X-API-Key: $API_KEY"

​
Configuring Thresholds

ThresholdRecommended Action
90-100Full autonomy — agent operates without intervention
70-89Standard operation — normal monitoring
50-69Enhanced monitoring — alert team, increase logging verbosity
30-49Restricted mode — require human approval for critical actions
0-29Automatic suspension — kill switch triggered

​
Witness Network

Agents can participate in a witness network where peers attest to each other’s behavior.

​
Create a Witness Request

curl -X POST https://api.truthlocks.com/v1/witness \
  -H "X-API-Key: $API_KEY" \
  -d '{
    "subject_agent_id": "maip-agent:01JXXXX",
    "claim": "Completed invoice processing without errors",
    "evidence": { "task_id": "task_456", "records_processed": 150 }
  }'

​
Attest as Witness

curl -X POST https://api.truthlocks.com/v1/witness/maip-witness:01JWWWW/attest \
  -H "X-API-Key: $API_KEY" \
  -d '{
    "attestor_agent_id": "maip-agent:01JBBBB",
    "verdict": "confirmed",
    "confidence": 0.95
  }'

​
Consensus

curl -X POST https://api.truthlocks.com/v1/witness/maip-witness:01JWWWW/consensus \
  -H "X-API-Key: $API_KEY"

{
  "witness_id": "maip-witness:01JWWWW",
  "consensus": "confirmed",
  "confidence": 0.91,
  "attestations": 4,
  "quorum_met": true
}

​
Integration with Risk Engine

Trust scores feed directly into the Truthlocks risk evaluation engine. Low trust scores automatically increase the risk signal for any action the agent takes: 
Trust Score < 50 → Risk Signal: HIGH → Action blocked or requires approval

​
Usage metering

Each on-demand trust score computation increments the maip.trust_computes usage counter for your billing cycle. Background score updates triggered by the platform (such as after agent actions) also count toward this metric. You can monitor your consumption with the usage API or in the console at Settings > Billing > Usage under the MAIP section.
Trust score compute quotas vary by plan tier. See the billing overview for details on all MAIP metered products.

​
Next steps

Machine Identity

Full overview of the Machine Agent Identity Protocol.

Cross-Tenant Delegation

Enable agents to operate across organizational boundaries.