The organisation that deploys the agent, in the end. No single official body certifies AI agents for use in a given business. Tests can be run by a platform, the vendor or an outside firm, but the deployer chooses the cases, sets the floors, and has a named, independent person accept or reject the result. That acceptance is the certification.
Is there an official certifier?
Not for individual agents in the sense a business needs. Several things exist nearby:
- ISO/IEC 42001 certification, issued by accredited certification bodies, covers an organisation's AI management system, not a specific agent's behaviour.
- Third-party standards for AI agents, such as AIUC-1 from the Artificial Intelligence Underwriting Company, set requirements a provider can be assessed against.
- Regulation, such as the EU AI Act, sets conformity requirements for some high-risk AI systems, which fall mainly on providers.
Each has a role. None tells a particular bank, insurer or utility that a particular agent may handle its customers' cases at a particular depth. That question can only be answered with the organisation's own evidence.
Who does what in certification?
Accountability is spread across a few roles, each recorded.
| Role | Responsibility | What is recorded |
|---|---|---|
| Process owner | Defines the capability and the depth sought; sets floors with risk | Capability definition, floors and the date set |
| Test operator | Runs the agent against the case set and scores it | Case set version, configuration, results per gate |
| Subject experts | Supply hard cases and agree expected outcomes | Cases and expected outcomes |
| Independent reviewer | Accepts or rejects the evidence; sets conditions | Named acceptance, conditions, date |
| Risk or second line | Oversees the method and challenges results | Review notes, challenges and how they were closed |
| Platform | Enforces the certified depth at the moment of action | Every action linked to the certificate in force |
The test operator can be internal, the agent's vendor or an outside firm. The reviewer should not be anyone who built the agent, bought it, or will be judged on its going live.
Why must the reviewer be independent?
Because pressure to pass is real. The team that built the agent wants it to work. The team that bought it wants the purchase justified. The business wants the savings. Each will tend, in good faith, to read results generously.
An independent reviewer reads the failures, not just the totals. A high average with a single failure on a vulnerable customer is a fail, and someone has to be willing to say so; see an average score hides the failure that matters.
How is accountability recorded?
On the certificate and in the record of actions. The certificate names the reviewer, the date and the conditions. Each action the agent takes is filed with a reference to the certificate in force. If something goes wrong, the trail shows who accepted which evidence and when.
This matters to supervisors. In regimes that expect named individual accountability, a certified agent working in a business area sits under the person accountable for that area. The certificate shows they had evidence before allowing the agent to act.
An illustrative example
Take an insurer deploying an agent to request missing documents on new claims. The vendor runs the tests using the insurer's past claims, through a testing platform. The claims operations manager defines the capability and, with the risk team, sets the floors. Two senior claims handlers add hard cases and agree the expected outcomes.
Results come back passing every gate except escalation on claims mentioning a death. The vendor proposes adjusting the expected outcome. The reviewer, the head of claims quality, who was not involved in selecting the vendor, rejects that and asks for a brief fix. The rerun passes. She accepts the certificate with a condition that bereavement claims always go to a person. Her name and the condition are on the certificate, and every request the agent sends is filed against it.
What should the organisation not hand over?
The choice of cases, the floors and the acceptance. A vendor or outside firm can help with all three, and often adds rigour, but the decision to let an agent act on your customers rests with you. The steps are in how to certify an AI agent, and whether certification is possible at all is answered in whether an AI agent can be certified.
Where Pryme fits
Attest runs the testing on the customer's own cases and scores seven gates against floors. The customer's named reviewer accepts the result, and Pryme enforces the certified depth at the moment of action and files every verdict. The wider reasoning is in what it means to certify an agent.
Questions people ask
Can an auditor certify an agent?
An auditor can review your certification process and test its results, which adds assurance. Auditors generally avoid making the decision themselves, because that would compromise their independence.
Does a regulator certify AI agents?
Not individually in most markets. Regulators supervise how firms govern their systems and may ask to see certification evidence.
Can certification be done by a committee?
A committee can approve the overall approach. Accepting a specific certificate works better with a named individual who reads the results and answers for the decision.