Why AI in PV Is Not Optional Anymore

Every module in this course from Module 4 onward has mentioned AI somewhere — a coding-assistance lesson here, a narrative-drafting caveat there, a causality-suggestion warning in Module 6. That distributed treatment was deliberate: AI genuinely does touch every one of those activities, and learning it alongside the underlying skill makes the human-in-the-loop principle concrete rather than abstract. This module exists because AI in pharmacovigilance has also become a coherent subject in its own right, with its own taxonomy, its own named production platforms, its own governance literature, and its own career implications — substantial enough now to deserve the same depth treatment this course has given MedDRA coding, causality assessment, or aggregate reporting.

The scale numbers explain why this shift happened as fast as it did. VigiBase, the WHO-UMC global database covered in Module 7, holds more than 32 million individual case safety reports. FAERS processes over 3 million reports annually. EudraVigilance receives hundreds of thousands more. PubMed adds roughly 1.3 million new entries a year, and social media generates millions of health-related posts daily. No manual system — no matter how well-staffed — genuinely covers that volume, across that many languages and sources, in anything approaching real time. The honest framing isn’t "AI or no AI." It’s "AI with appropriate governance, or data overload with inadequate coverage."

The market has already made this decision at scale. The global PV market exceeded USD 10 billion in 2025; the AI-in-PV segment specifically, worth roughly USD 600 million in 2024, is projected to approach USD 2 billion by 2034 at a compound annual growth rate above 20%. By mid-2025, six of the top 25 global pharmaceutical companies had selected AI-driven case processing platforms as their primary PV system — not a pilot, not a proof of concept, the platform their case processing actually runs on. That adoption has only accelerated since: by 2026, agentic AI specifically has become what industry analysts call the defining competitive battleground among the major safety database vendors.

This module works through that reality in the same order the rest of this course has used for every complex subject: first the taxonomy (what "AI" actually means, since it covers wildly different techniques with wildly different governance needs), then what it does at each stage of the workflow this course has already taught, then the specific named platforms running in production today, then the governance framework — CIOMS WG XIV above all — that makes deploying any of it defensible, and finally the human-in-the-loop discipline that determines whether an AI-augmented PV system is actually safer or just faster.