The Money Is Ahead of the Rulebook in Trial AI

The gap between forecast and framework

Two numbers should make life sciences leaders uneasy at the same time. Generative AI in clinical trials is projected to reach $1.98 trillion by 2035, according to SNS Insider. High-risk IMDRF Category III and IV software as a medical device is growing at a 14.6% CAGR in the US, an expansion analysts attribute directly to an established FDA pathway covering lifecycle management, cybersecurity, and predetermined change control plans, per Fact.MR. Capital is treating clinical trial AI as a mature, de-risked category.

It is not. An investment bank analysis reported by BioXconomy found that the primary barrier to AI adoption among clinical trial sponsors and CROs is not technical readiness. It is the absence of explicit regulation. Bourne’s head of research, Donald Hooker, told the outlet plainly that today there is limited explicit regulation of AI used in trial conduct itself.

Two different categories, one confused market

The confusion is that “FDA has a pathway for AI” and “AI used to run a trial is regulated” are not the same statement. The Fact.MR growth figures describe AI as a medical device, a diagnostic or therapeutic product subject to premarket review, real-world performance monitoring, and the FDA’s own predetermined change control framework. That pathway governs the product. It says nothing about the AI a sponsor uses internally to design protocols, match patients, monitor sites, or clean data during the trial that generates the evidence for that product’s own approval.

This second category, AI embedded in trial execution rather than in the device itself, is where the regulatory vacuum sits. It matters because execution is still the bottleneck. As TruTechnologies’ Richard Graham has noted, no AI-discovered drug has yet cleared the full trial process, and the reason has less to do with molecule design than with the point-of-care data and execution mechanics that AI-discovery speed does nothing to fix, per Applied Clinical Trials. Sponsors are automating the part of the process regulators have not yet defined rules for, in service of the part regulators have.

The compounding problem for global sponsors

For sponsors running multi-region trials, this uncertainty stacks. The EU AI Act, in force since August 2024, is rolling out high-risk obligations in stages, with the next major tranche not landing until December 2027, per Diplomacy and Law. Whether trial-conduct AI tools eventually fall into a high-risk category under that framework is still an open interpretive question. A sponsor building a validation and documentation regime today has no single authoritative target on either side of the Atlantic.

What to actually do about it

The sound move is not to wait for FDA or the EU to draw the line. It is to govern trial-conduct AI to the standard regulators will eventually demand, using ISO 42001 as the internal scaffold: documented validation, bias testing against trial populations, audit trails for every AI-assisted decision that touches enrollment, dosing, or data integrity. Absent that, a sponsor’s trial-conduct AI becomes the weakest link in the evidence chain, discoverable by any regulator auditing the very approval it helped produce.

The forecasts assume the rules will catch up. Until they do, governance is the only thing standing between a fast trial and an indefensible one.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The core argument—that capital is flowing into trial-conduct AI faster than regulation can govern it—is coherent and the device-vs-execution distinction is valid, but the $1.98 trillion figure appears
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but some sources could be more authoritative or directly relevant to the claims they support.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects the regulatory gaps and standards (ISO 42001, EU AI Act, FDA pathways) but slightly understates the FDA’s emerging guidance on AI in trial execution (e.g., draft frame
Technical Accuracy	Llama	cleared. The article accurately distinguishes between AI used as a medical device and AI used in trial conduct, highlighting a regulatory gap that is a critical concern for clinical trial sponsors.
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and counters potential vendor hype by distinguishing between regulated AI as a medical device and unregulated AI in trial conduct, using specific market data and ex
Novelty & Non-Duplication	Grok	held. Core thesis and figures are straight lifts from the BioXconomy regulatory-uncertainty piece plus SNS/Fact.MR releases; the SaMD-vs-trial-conduct distinction is only incremental framing, not a new find
Validation	DeepSeek	cleared. The central claim that a regulatory vacuum exists for AI used in trial execution is validated by direct expert testimony and the absence of cited regulations to the contrary.

Sources cited: 14. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.