Know what your agents are doing. Continuously.

Governance signals from every agent platform.

Trust3 AI connects to your agent platforms and ingests telemetry continuously. That raw telemetry is aggregated into governance metrics — the signals that matter for monitoring, scoring, and violation detection.

• Interaction volume — total interactions per agent, per time window
• Scope categorization — subject matter of agent interactions, classified against declared purpose
• In-scope vs out-of-scope ratio — what proportion of usage matches the agent's declared purpose
• Data access patterns — which resources agents are reaching and how frequently
• Policy evaluation results — which policies fired, with what outcomes
• Success and failure rates — completed, failed, and flagged interactions

• Databricks — Unity Catalog connector, CloudTrail, Genie space activity
• Microsoft Copilot Studio — M365 Unified Audit Log
• AWS Bedrock — CloudTrail and Bedrock audit logs
• LangChain / custom agents — PAIG SDK telemetry
• Any platform emitting OpenTelemetry-compatible data

A live governance grade. Per agent. Computed continuously.

Every agent carries a Trust Score — a 0–100 grade across four dimensions that reflects its current governance posture. Computed continuously from aggregated metrics as new telemetry arrives. Not a static badge assigned at registration — a live signal that moves with the agent's actual behavior.

• TRUSTED All policies met, operating within declared scope
• MONITORED Minor signals detected, within acceptable range
• AT RISK Active violations or significant drift detected
• UNTRUSTED Critical violations, production risk

Drilling into a score shows exactly which dimension is the primary driver and which specific signal is pulling it down — policy violation count, out-of-scope ratio, or behavioral deviation.

Understand what your agents are actually doing.

Aggregated usage metrics give governance teams a continuous view of how each agent is being used — and whether that usage matches what the agent was built for.

• Total interactions — daily, weekly, rolling 30 days
• In-scope vs out-of-scope breakdown — categorized against declared purpose
• Subject category distribution — the topics agents are actually handling
• Success rate — completed, failed, and flagged interactions
• User and team breakdown — which principals are driving volume

When aggregated telemetry shows interactions falling outside an agent's declared scope, Trust3 AI categorizes the deviation and flags it as a policy violation. The violation record includes the category of out-of-scope usage and a reference to the source system for detailed investigation.

Example: A sales data agent's telemetry shows interactions categorized as weather queries and HR/benefits topics — both outside its declared sales analytics scope. Trust3 AI flags the category, creates issues, and points to the originating platform for the specific interaction detail.

Catch problems from aggregated signals.

Trust3 AI runs continuous analysis across incoming aggregated metrics — detecting patterns that indicate violations, drift, or governance risk.

• Out-of-scope usage — interactions categorized outside the agent's declared purpose, detected from subject category metrics
• Scope drift — gradual shift in an agent's usage category distribution over time, detected against its established baseline
• Policy violations — access to data or resources that active policies prohibit, detected from access pattern metrics
• Behavioral anomalies — usage patterns that deviate significantly from the agent's established baseline
• Data access violations — agents reaching tagged data assets (PII, PHI, PCI) outside declared scope

Sensitivity is controlled with a single slider. Security and compliance teams set the threshold. Trust3 AI handles the continuous analysis.

Every violation becomes a traceable, assignable record.

When monitoring detects a policy violation or scope deviation, Trust3 AI automatically creates an issue — timestamped, categorized, and linked to the agent and the active policy. No manual triage. No spreadsheets.

• The agent that triggered the violation
• The policy that fired and the violation category
• A reference link to the source system for detailed trace investigation
• Severity level: CRITICAL, HIGH, MEDIUM, LOW
• Workflow status: Detected → Under Review → Remediated → Resolved

Issues are assignable, trackable, and retained permanently as your compliance record. The violation category and governance context are what regulators and audit teams need. The full trace detail is in your source system, accessible via the evidence link.