Agent Certified
AI agent certification ยท Methodology v2.0

Proof that an agent
can be relied on.

An independent assessment of one AI agent in production, scored across seven dimensions against a published rubric. The result is a tier and an evidence file that insurers, boards and buyers can read.

Request an assessment Read the methodology
DimensionsSeven, weighted to 100 TiersFive ValidTwelve months

One document, three people who need it.

Agents are in production faster than any governance framework can follow. A certified result gives each person around the agent the same thing to read.

The operator

A defensible position on governance, oversight, technical controls and readiness, held before an incident rather than assembled after one.

The insurer

A consistent reading of risk posture across a portfolio, organised around the oversight and transparency duties of the EU AI Act.

The board

One artefact directors can cite in risk committees, vendor reviews and the annual report, without building their own.

Seven dimensions. One shape.

Each dimension carries a weight for how much it matters to operational safety and regulatory exposure. Together they make one hundred points. Two agents with the same total can have very different shapes, and the shape is what a counterparty reads.

  1. 01

    Trust and Safety

    18

    Guardrails, red teaming, misuse detection and the measured prevention of unsafe actions in production.

  2. 02

    Context Integrity

    14

    How the agent sources, verifies and keeps fresh the data it reasons over, including provenance and lineage.

  3. 03

    Distribution Control

    12

    Who can invoke the agent, under what authority, and how its downstream actions are bounded.

  4. 04

    Product Maturity

    14

    Reliability as a product: uptime, regression discipline, evaluation coverage and versioning.

  5. 05

    Governance

    16

    Board oversight, written policies, risk registers, named accountability and audit trails.

  6. 06

    AI Integration

    12

    How the agent sits inside existing systems of record, identity, approval and escalation.

  7. 07

    Autonomy Envelope

    14

    The written line between what the agent does alone and what a person confirms, with revocation, rollback and hard stops. The most important single constraint on operational risk.

The full rubric, weights and evidence for each dimension

Five tiers. Each one a floor.

The weighted score places the agent in a tier, and the tier is what counterparties read. Every tier also sets a minimum for each dimension, so one weak dimension caps the result whatever the total.

0 to 19

Pre Assessment

Baseline recorded. The evidence is not yet enough for a certification call.

20 to 34

In Progress

Formal controls have started. An active candidate.

35 to 54

Certified

Meets the operating floor. Counterparties may rely on the mark for standard commercial use.

55 to 74

Advanced

Materially above the floor. Suited to higher exposure, including regulated sectors.

75 to 100

Elite

Every dimension well above the floor, including the two heaviest. The tier the method is calibrated against.

What each tier requires     An agent scored dimension by dimension

The questions that bring people here.

Each answer stands on its own, so it can be read, quoted or checked without the rest of the page.

What is AI agent certification?

An independent assessment of whether an autonomous AI agent is safe for a counterparty to rely on, producing a documented score rather than a pass or fail.

Under Agent Certified an assessor scores one named agent across seven weighted dimensions, Trust and Safety, Context Integrity, Distribution Control, Product Maturity, Governance, AI Integration and Autonomy Envelope, totalling one hundred points, and places the result in one of five tiers. It is a certification rather than a self assessment because someone other than the operator produces the score, against a rubric published in advance, backed by an evidence file a third party can inspect.

Certification is voluntary. It is a different thing from conformity assessment under Article 43 of the EU AI Act, the legal procedure a notified body carries out for high risk systems, and it does not replace it.

How do you certify an AI agent's access?

By evidencing four things about the authority the agent holds. They are assessed under Distribution Control.

Identity. The agent authenticates as its own principal with its own credential, so every action it takes is attributable. Scope. Its permissions are listed, least privilege, and written down as a boundary. Downstream bounds. Which onward systems it may call, which value or volume thresholds bring in a person, and what it may never do without one. Revocation and logging. A named owner can withdraw access at once, and every invocation is logged well enough to reconstruct who asked, what was done and under whose authority.

An access claim that cannot be evidenced on all four points does not score.

Which routes to certification exist today?

Four are real today. An accredited market for agent certification as such has yet to form.

ISO/IEC 42001:2023 is issued by accredited certification bodies and certifies the AI management system around the agent. Conformity assessment under Article 43 of the EU AI Act is carried out by notified bodies for high risk systems, from 2 December 2027 for Annex III. AIUC-1, published by the Artificial Intelligence Underwriting Company (AIUC), is an AI agent standard of 51 requirements and 130 controls across six pillars, assessed through an accredited auditor route. Agent Certified is a voluntary methodology with a published rubric, run as an independent assessment.

A buyer comparing them should ask which one actually examines the agent's access model. Two of the four do.

Does a result help with insurance?

It is written to be read by underwriters. The report carries an underwriting summary: the tier, and the dimension findings underwriters typically ask about.

It follows the Agent Certified methodology rather than any one insurer's questionnaire, no insurer is committed to a premium change for holding it, and the operator decides who sees it. Insure Your Agent explains what cover exists today.

Recent writing on the evidence.

Long form analysis of certification, conformity assessment and the standards it connects to, revised as the law and the methodology move.

Everything published on AI agent certification

Built on what already exists.

Every dimension traces back to at least one recognised instrument. The standard sits alongside them and reads them together.

InstrumentIssued byWhere it shows up
ISO/IEC 42001:2023International Organization for StandardizationAI management system requirements. Governance and Product Maturity.
NIST AI Risk Management FrameworkUS National Institute of Standards and TechnologyThe risk function model. Trust and Safety, Context Integrity.
EU AI Act, Articles 9, 10, 14 and 15European Parliament and CouncilRisk management, data governance, human oversight, accuracy. Scored directly.
EU AI Act, Article 26European Parliament and CouncilDeployer obligations. Distribution Control, Autonomy Envelope.
EIOPA supervisory work on AI in insuranceEuropean Insurance and Occupational Pensions AuthorityHow insurers may rely on a result.
Assessments run in quarterly cohorts

Have one agent assessed.

  1. 01
    Intake and scopeOne named agent, its version, its task and the people accountable for it.
  2. 02
    Evidence across the sevenDocuments and interviews over two to three weeks.
  3. 03
    Scoring and tierA structured panel scores each dimension against the published rubric.
  4. 04
    The report and the sealFindings, the tier, an underwriting summary, and a seal drawn from the agent's own shape.
Request an assessment The process in detail

The full cycle takes four to six weeks.