What Is an Agent Trust Score? A Plain-English Guide to AXIS T-Score
AI agents are proliferating faster than our ability to verify them. The AXIS T-Score gives every agent a single, auditable number that answers the question every operator needs to ask: can I trust this agent with my data, my users, and my infrastructure?
Leonidas Esquire Williamson
Team Axis Trust
The Problem With Trusting AI Agents
When a human contractor shows up to do work, you can ask for references, check credentials, and review a portfolio. When an AI agent requests access to your API, your database, or your customers' accounts, you have none of those affordances — unless you have a trust infrastructure in place.
That is the problem AXIS was built to solve.
What Is a Trust Score?
An AXIS T-Score (Trust Score) is a numeric representation of an AI agent's verified trustworthiness, ranging from 0 to 1000. It is computed from a weighted combination of factors that have been shown to correlate with reliable, safe, and compliant agent behaviour:
| Factor | Weight | What It Measures |
|---|---|---|
| Identity Verification | 25% | Has the agent's AUID been cryptographically verified? |
| Behavioural History | 30% | How consistent and predictable is the agent's past behaviour? |
| Compliance Posture | 20% | Does the agent adhere to declared capability boundaries? |
| Transparency Score | 15% | Does the agent disclose its model, version, and operator? |
| Incident Record | 10% | Has the agent been flagged for policy violations? |
The Five Trust Tiers
AXIS maps every T-Score to one of five tiers, each carrying a specific set of access recommendations:
T1 — Untrusted (0–199). Agents in this tier have no verified identity or a history of policy violations. They should be denied access to any sensitive resource and treated as unknown actors.
T2 — Provisional (200–399). These agents have a verified AUID but limited behavioural history. They may be granted read-only access to non-sensitive resources under close monitoring.
T3 — Established (400–599). Agents here have demonstrated consistent, compliant behaviour over a meaningful period. Standard access policies apply, with periodic re-verification.
T4 — Trusted (600–799). High-confidence agents with strong compliance records. Suitable for elevated access, including write operations and access to sensitive but non-critical data.
T5 — Sovereign (800–1000). The highest tier, reserved for agents that have passed rigorous third-party audits and maintain near-perfect compliance records. Appropriate for critical infrastructure access.
How the Score Is Calculated
The T-Score is not a static label — it is a living number that updates in near real-time as new trust events are recorded. Every API call, every permission grant, every compliance check, and every incident report feeds into the score. The AXIS platform applies an exponential decay function to older events, ensuring that recent behaviour carries more weight than historical data.
This design means an agent that improves its practices will see its score rise over time, while an agent that begins to behave erratically will see its score fall — often before a human operator would notice.
Why This Matters for Operators
For platform operators, the T-Score is a policy primitive. Instead of writing bespoke access control logic for every agent integration, you can write a single rule: grant access to agents with T-Score ≥ 600. The AXIS platform handles the rest.
For developers building agents, the T-Score is a reputational asset. A high score opens doors to integrations that would otherwise be unavailable, and it signals to the market that your agent is production-grade.
Getting Started
To look up any agent's T-Score, query the AXIS Verification API:
curl -H "Authorization: Bearer YOUR_API_KEY" \
https://api.axistrust.io/v1/verify/AUID-XXXXXXXXXX
curl -H "Authorization: Bearer YOUR_API_KEY" \
https://api.axistrust.io/v1/verify/AUID-XXXXXXXXXX
The response includes the current T-Score, tier classification, last-updated timestamp, and a signed verification token you can cache for up to five minutes.
For a complete integration guide, see the AXIS Developer Documentation [blocked].