Can an AI agent track compliance training?

Yes. Training records are among the simplest compliance jobs to hand over. An AI agent can work out who needs which training from their role, track completion, chase people who are overdue, trigger refreshers when policies change and report to management. Granting exemptions, deciding consequences for persistent non-completion and judging whether the training actually works stay with a person.

What does tracking training involve?

Regulated firms must make sure staff have the knowledge their roles need: AML for anyone handling customers or payments, conduct rules for those dealing with customers, data protection for anyone handling personal data, market abuse for trading staff, and so on. Under the EU AI Act, Article 4 adds an expectation of AI literacy for staff dealing with AI systems.

The tracking work is mostly bookkeeping. Who joined, who moved roles, who is due a refresher, who has not finished, whose manager needs to know. It falls between HR and compliance and is often run from spreadsheets that are out of date the day they are made. An agent can keep it current.

What may the agent prepare?

The agent may:

• Map each role to its required training, from the training policy.
• Assign training to new joiners and to people changing roles.
• Track completion and send reminders on a schedule, escalating to managers when overdue.
• Trigger refresher training when a relevant policy or regulation changes; see can an AI agent track regulatory change?
• Flag people in high-risk roles whose training has lapsed.
• Report completion by team, role and module to compliance and management.
• Analyse assessment results to show which questions people get wrong most.

People decide:

• Which training each role requires, and how often.
• Exemptions, such as for long-term leave.
• Consequences for persistent non-completion, including effects on fitness and propriety or bonus.
• Whether the content is adequate, and whether to change it.
• Whether someone whose training has lapsed should be stopped from certain work until it is done.

The last decision is often overlooked. A trader without current market abuse training, or a payments officer without current AML training, is a risk the firm should not carry quietly. The agent flags it; a person decides whether to restrict the role.

What does tracking look like in practice?

Take a mid-sized bank with 1,500 staff. The agent tracks seven compliance modules.

In one month it assigns training to 40 new joiners and 25 people who moved roles. It sends reminders to 180 overdue staff and escalates 30 to their managers. It flags six people in the payments team whose AML training lapsed over two months ago. It also notices that after a change to the bank's complaints policy, nobody assigned a refresher, and that a question about complaint handling time limits is answered wrongly by half of those who sat the complaints module.

The head of compliance restricts the six payments staff from approving payments until they complete. She asks the learning team to rewrite the complaints module section on time limits, and approves a refresher for customer-facing staff. The agent tracks the remediation through to completion, much as it would track any remediation action.

Does tracking training apply to agents too?

The same question arises for AI agents: how do you know an agent knows what it needs to? Our answer is that agents are qualified, not trained. We do not retrain model weights for each firm; we change an agent's brief, sources and bounds, and certify it on the firm's own cases. The full argument is in qualification, not training.

People who approve agents' work do need training, though: they must understand what the agent did, what it could not see, and what their approval means. That point is made in approvers need training too, and the agent tracking training should include those approvers in its scope.

Which controls make tracking reliable?

• Role mapping comes from HR records. The agent uses the system of record for roles and joiners, not a copy.
• Policy defines requirements. The agent applies the training policy; it does not decide who needs what.
• Escalation is fair. Reminders follow a set schedule, and escalation to managers is the same for everyone.
• Completion evidence is kept. Each completion links to the learning system's record.

The general principle is in our guide to AI compliance agents.