Agent Authorization
Every AI agent in the Truthlocks platform operates under a strict authorization model. Agents must be registered, assigned scopes, and create time-bounded sessions before they can take any action.Authorization Flow
Scope model
Scopes follow a hierarchicalresource:action pattern. When an agent requests a session, it can only request scopes that were assigned at registration.
Defining Custom Scopes
Built-in Scopes
| Scope | Description |
|---|---|
agents:read | List and inspect agents |
agents:write | Register and update agents |
receipts:write | Mint action receipts |
attestations:read | Read attestation data |
attestations:write | Create new attestations |
trust-scores:read | Read trust score data |
trust-scores:compute | Trigger trust score recomputation |
delegations:offer | Offer cross-tenant delegation |
delegations:accept | Accept delegation offers |
compliance:write | Create compliance checks |
orchestrations:execute | Execute multi-agent workflows |
datasets:attest | Attest dataset provenance |
models:attest | Attest model lineage |
Session Management
Sessions provide time-bounded execution contexts with automatic expiry and optional IP allowlisting.Creating a Session
Session Lifecycle
| Status | Description |
|---|---|
active | Session is valid and accepting requests |
expired | TTL elapsed — automatic cleanup |
terminated | Explicitly ended by user or system |
suspended | Paused due to anomaly detection |
Terminating a Session
Tool Registration & Invocation
Agents can register tools they expose and invoke tools registered by other agents (subject to scope checks).Register a Tool
Invoke a Tool
- Scope-checked — the session must have the required scope
- Rate-limited — per your plan’s tool invocation quota
- Audit-logged — full input/output recorded in the audit trail
- Metered — counted against your billing entitlements
Security Best Practices
- Principle of least privilege — Request only the scopes needed for the current task
- Short-lived sessions — Use the shortest practical TTL (default: 1 hour)
- IP allowlisting — Restrict sessions to known network ranges
- Rotate credentials — Regenerate agent signing keys on a regular schedule
- Monitor trust scores — Set alerts when trust scores drop below thresholds
- Use the kill switch — Immediately revoke compromised agents
Next steps
MAIP policies
Runtime enforcement rules based on trust scores, scopes, and delegation depth.
Trust Scores
Continuous behavioral trust evaluation for agents.
Cross-Tenant Delegation
Allow agents to act across organizational boundaries.

