AI Trust Gap Slows Down Innovation in Clinical Research

rtifical Intelligence is poised to confer significant benefits in all areas of clinical research: protocol design, site selection, patient recruitment and enrollment and engagement, insights and decision making, monitoring, and medical writing.

There is an AI trust gap in the clinical research industry, as we work to demonstrate the utility of new and non-deterministic models in a highly regulated industry. The AI trust gap will hold back the realization of this value.

Trust erosion manifests as operational consequences of premature deployment: sites refusing tools, ethics committees delaying approvals, and sponsors quietly reverting to manual workarounds. As negative experiences by early adopters are more prominently publicized, trust erodes, and development interest and adoption wanes.

AI offers incredible potential to improve outcomes in clinical trials. As an industry, we need to partner to build trust in responsible use of AI.

A Tradition of Publication and Transparency

The life science industry has established mature and trusted publication and transparency frameworks (see below). Our rigorous existing norms should apply to AI development and deployment. This process can accelerate trust and adoption, to leverage the power of AI to improve outcomes in clinical trials, faster. Given the adoption of transparency frameworks in all aspects of biomedical research, we should apply similar principles to sharing experiences gained with AI initiatives geared toward the greater good of getting medical advancements to patients quicker, cheaper, and with higher quality. This publication and transparency framework requires us to report both positive and negative findings in our use of AI in clinical research.

We ask the clinical research industry to create a registry of AI outcomes in clinical trials to enable our industry to report on their use of AI to accelerate positive outcomes.

Apply Existing Trust Frameworks to AI

Clinical research is bound by formal, and in many cases legally enforceable, frameworks that require researchers to report both positive and negative outcomes: Good Clinical Practice (GCP), trial registration and results-reporting mandates such as FDAAA 801, International Committee of Medical Journal Editors Guidelines (ICMJE), and CONSORT Guidelines. This is what distinguishes clinical research from most other stages of science: the reporting of both positive and negative outcomes isn’t just encouraged, it’s required. Additionally, under the Declaration of Helsinki, researchers have an ethical obligation to make positive, inconclusive, and negative results publicly available.

When publishing research, we not only focus on positive outcomes but also strive to provide a fuller account, including negative or unexpected effects, protocol deviations, and self-critique of research design. We do this because healthcare providers, patients, payers, and policy makers deserve a complete picture for critical decisions affecting millions of patients. Disclosing a complete picture, rather than a selectively favorable narrative, is not only the right thing to do, but it also builds trust in our products and services. When bad actors violate this trust, it negatively affects good actors.

As an industry, we should follow these same principles to report AI outcomes, both positive and negative. Scientific publications and data sharing come with the understanding that the information may also be used by competitors to achieve the greater good in a better and faster way. Under these principles, we can accelerate the use of AI that helps us advance medicine.

Apply Existing Norms and Principles to AI

Publish Results to Benefit the Clinical Research Industry
The dissemination of clinical findings, both positive and negative, regardless of commercial intent, results in healthier and happier lives and is considered an ethical responsibility. While there is no shortage of press releases and conference presentations publicizing AI successes, shared results must adhere to our industry’s standard of rigor.

The Coalition for Health AI defines accountability as the responsibility of those involved in AI development and deployment to maintain auditability, minimize harm, report negative impact, and communicate design trade-offs. This transparency-driven accountability builds trust with key stakeholders, helps others understand the full context of findings for more responsible use, and helps others learn from mistakes to minimize or prevent the prolonging of human suffering.

Publish More AI Outcomes in Clinical Trials
The clinical research industry needs more knowledge of both AI successes and failures. We call upon our industry to create a registry of AI outcomes in clinical trials to enable our industry to report on their use of AI to accelerate innovation.

There may be hesitancy in sharing AI outcomes because the regulatory treatment of AI in clinical trials is still unsettled. Sponsors may worry that voluntarily disclosing an AI-related failure could be misread as noncompliance or invite scrutiny from regulators who haven’t yet clarified how they’ll treat AI use in trials. And currently many organizations hesitate to share true success stories (or failure stories) publicly, to preserve competitive advantage.

Life science professionals recognize the disclosure of shortcomings not as a negative, but as an opportunity for growth, intellectual honesty, and building public trust. Common shortcomings in clinical research are directly analogous to those in AI implementations: limitations in data sampled, selection bias, methodological flaws, and constraints on generalizability. If we are serious about accelerating trust, we must name the successes and failures people are already experiencing but not sharing.

Call to Action

Transparency in publishing clinical research results is not simply a matter of goodwill. Trial registration and results reporting are mandated under previously mentioned frameworks like FDAAA 801, ICMJE, and GCP, which is why negative and inconclusive findings reach the public record even when they are commercially inconvenient. That mandate does not extend upstream. In preclinical and basic research, no comparable reporting requirement exists, and negative results are known to go unpublished. AI in clinical trials sits in this same unregulated space today; there is no requirement to report what fails. If we want AI development to earn the trust that clinical research has earned, we need to build the equivalent of that mandate voluntarily, starting with a registry of AI outcomes, positive and negative.

This registry cannot credibly sit inside any single sponsor, technology vendor, or service provider. The same conflict-of-interest logic that makes precompetitive data-sharing valuable also applies to who is trusted to hold it. What it needs is a steward that meets three criteria. It must have standing across sponsors, CROs, sites, and technology vendors—not allegiance to one. Its governance must be built for transparency and accountability, along the framework that the Coalition for Health AI has already broadly outlined for AI development. And it must have no commercial stake in the outcomes being reported. Our industry already has precompetitive infrastructure built for exactly this kind of shared, noncompetitive problem. Bodies like the Clinical Trials Transformation Initiative, TransCelerate BioPharma, and the Clinical Data Interchange Standards Consortium (CDISC) exist because competitors recognized that trial methodology, data standards, and biomarker science aren’t worth guarding competitively. That’s the logical foundation to build on.

Our interdependency in this is embodied in Rudyard Kipling’s observation that “the strength of the pack is the wolf, and the strength of the wolf is the pack.” By applying transparency norms, we can accelerate trust in AI and raise the tide that lifts all life science boats.