Overview

Agentic AI has arrived in pharmacovigilance. Systems that read safety correspondence, build cases, code events and draft assessments are moving from pilots into production, and the job of the pharmacovigilance professional is changing with them: from executing the work to supervising the machines that execute it. Supervision is a different skill from execution, and almost nobody in the field has been trained for it.

This virtual live training course gives pharmacovigilance professionals the vocabulary, the mental model and the practical tools to oversee AI agents competently and compliantly.

Participants learn how an agent actually works, where it fails and why those failures are hard to see, how to review agent output with a defensible sampling and escalation plan, how to apply credibility assessment and validation thinking to a non-deterministic system, and what has to change in SOPs, roles and KPIs when the work is delegated.

The course is hands-on. Every participant works in a browser-based sandbox with a representative ICSR intake agent, breaks it, supervises it and writes its credibility plan. It is taught by a physician with direct experience of building and operating agentic pharmacovigilance systems and of defending them in front of auditors and inspectors. EU and US regulatory expectations are treated side by side throughout.

Featured topics

- Agent anatomy: the loop, tools, memory, traces and non-determinism
- Where agents fit in the PV value chain and where they fail
- Human-in-the-loop, on-the-loop and out-of-the-loop supervision, human-in-command concept
- Risk-based sampling, confidence thresholds, escalation and audit trail
- Credibility assessment framework and validation of non-deterministic systems
- EMA and FDA expectations side by side: reflection paper, AI credibility framework, EU AI Act, CSA, Annex 11 and Part 11
- Agentic architecture patterns and where the human checkpoint belongs
- Change management: SOPs, roles, KPIs, training records and the psychology of delegation

Who should attend?

This virtual live training course is designed for Pharmacovigilance professionals whose roles are shifting towards oversight of AI-assisted or AI-executed processes:
- Case processors and case-processing leads
- Medical reviewers
- PV quality and compliance staff
- Local safety officers
- QPPVs and deputies
- PV system owners and business analysts

No technical background is required; a 20-minute pre-course primer brings participants with no prior AI exposure to the same starting point.

Learning objectives

After the completion of this virtual live training course, participants will be able to:
- Define agent, tool call, trace, guardrail, eval and golden set, and explain how an agent loop runs
- Classify pharmacovigilance tasks as delegate, delegate with checkpoint, or keep human, with reasons
- Recognise the pharmacovigilance-specific failure modes of AI agents, including silent omission, and name the control that catches each
- Review a batch of agent output using thresholds, risk-based sampling and escalation rules, and measure their own oversight quality
- Draft a context-of-use statement, model-risk rating and credibility assessment plan for one pharmacovigilance task, and map it to EU and US expectations
- Describe the changes to SOPs, roles, KPIs and training records that supervised-agent workflows require

Digital Learning Catalog

Download

Have an account?

Download registration form Register

Agentic AI in Pharmacovigilance: From Operator to Supervisor