Accreditation is formal recognition, by an authoritative body, that an organisation is competent to carry out a specific activity, such as assessing people or issuing certificates, to a defined standard. It judges the provider, not the learner. A certificate says a person met a standard; accreditation says the body that issued it was fit to judge.
Why does accreditation come up with AI agents?
Buyers and supervisors want to know whether the people overseeing agents are competent, and they reach for familiar signals. Accreditation is one of them. Clients increasingly ask suppliers what proof of training sits behind the staff who approve an agent's work, and the kinds of evidence they accept are set out in what evidence of training clients ask suppliers for.
The risk is treating the word as a substitute for the skill. Accreditation tells you a provider's processes were reviewed. It does not tell you whether a particular reviewer can tell a sound held action from a bad one this week.
How does accreditation work in practice?
An accrediting body sets criteria for providers: how assessments are designed, how assessors are qualified, how results are recorded, how appeals are handled, how quality is monitored. The provider submits evidence, is audited, and is reviewed on a cycle. If it falls short, it can lose the status.
For training in agent oversight, the useful questions sit underneath the label. How is competence assessed: a quiz, or real decisions on real cases? Who signs off? Does the certificate expire? Can a third party check its status at source? Those answers show the quality of a credential whether or not the provider is accredited.
What is an example?
Take an insurer choosing training for claims handlers who will approve an agent's settlement drafts. One provider is accredited and assesses with a single multiple-choice test that never expires. Another is not accredited but assesses with supervised review decisions on live cases, signed off by an assessor, with a certificate that lapses and can be checked by number.
The second tests what the job requires: meaningful human review of the agent's work. The first tests recall. Accreditation is worth weighing, but the method of assessment is what predicts behaviour at the review queue, which is why the case for certifying claims handlers before agents go live rests on assessed practice.
How does accreditation differ from certification?
Certification is awarded to a person (or, for agents, to a capability) after they meet a standard. Accreditation is awarded to the organisation that does the certifying. A certified reviewer is judged competent; an accredited provider is judged competent to judge.
There is a parallel for agents themselves. An agent's capability can be certified by replaying it against the customer's own cases, but that says nothing about whether the people approving its output are competent. Both layers need evidence.
Which terms are related?
- Certification: the award to the individual who met the standard.
- Assessor: the qualified person who judges the evidence.
- Rubber-stamping: the failure that good assessment is meant to prevent, explained in what rubber-stamping is.
For the whole topic, read AI literacy and certification for people who work with agents.
Questions people ask
Should we only accept accredited AI training?
Make it one factor, not the filter. Ask how competence is assessed, who signs off, whether the credential expires and whether you can check it at source, then weigh accreditation alongside those answers.
Can accreditation expire or be withdrawn?
Yes. Accrediting bodies review providers on a cycle and can suspend or withdraw recognition if standards slip. Check the provider's status at the time you rely on it, not only when you first signed up.