The speed-safety gap: pharma's least discussed AI problem

Three systems that have to keep pace

If AI compresses discovery and early trials, three things have to keep pace or the whole thing stalls.

Pharmacovigilance. Signal detection and case processing were architected for yesterday's volumes and cadences. Compress the front of the pipeline and the safety machinery downstream inherits a throughput problem it was never designed for: more candidates, faster cycles, and adverse-event signals that need catching at the new speed, not the old one.

Regulatory credibility. The FDA's 2025 draft guidance on AI in drug development is a risk-based framework for exactly this problem: a "credibility" standard built on transparency, validation and human oversight, with final guidance expected in 2026 and the EMA aligned in direction. The regulators are not slowing AI down; they are defining what trustworthy speed has to look like.

Data integrity. Faster models trained on incomplete or biased data do not just move fast. They move wrong, fast. Every compression of the timeline raises the price of a data problem, because there is less time for slow human review to catch what the pipeline missed.

Speed without governance is risk transfer

This is where a finance lens earns its place in a scientific conversation. Speed without a matching lift in safety and governance is not progress; it is risk transfer, from the development budget to the patient and the balance sheet. The cost has not disappeared. It has moved somewhere less visible and more expensive, into recall risk, litigation risk, and the regulatory trust that takes a decade to build and one failure to lose.

I spent years inside pharma finance watching how seriously the industry takes its control environment: the statutory disciplines, the quality systems, the pharmacovigilance obligations that sit behind every licence. That instinct, that the licence to operate is earned through governance, is precisely the asset the AI era needs. The companies that win this transition will invest in the guardrails as seriously as the horsepower.

Faster discovery is the easy half. Trustworthy at speed is the hard half, and the valuable one.

Sources: the FDA's 2025 draft guidance on AI in drug development (fda.gov/about-fda/center-drug-evaluation-and-research-cder/artificial-intelligence-drug-development); IQVIA on AI reshaping pharmacovigilance (iqvia.com/blogs/2025/09/how-ai-is-reshaping-pharmacovigilance).