FDA to speed up drug trials with AI and real-time data

The FDA is moving toward real-time clinical trials. The regulator wants to see key safety and efficacy signals while a study is still running, not after it ends. If the approach works, it could cut months from new drug development, and in some cases years.

How the process is changing

Today, clinical trial data usually travels down a long chain: from trial sites to the sponsor, then to analysis, and only after that to the FDA. This hurts most in early-stage trials. Patient numbers are small, uncertainty is high, and decisions on dosing, safety, and whether to continue a program often come with a noticeable delay.

The new approach targets exactly that lag. The regulator doesn't want the full mass of raw patient data. It wants key efficacy and safety signals almost immediately, while the study is still in progress.

To that end, the FDA announced two steps on April 28, 2026. The first is the launch of proof-of-concept trials that will send signals to the regulator in real time. The second is a request for proposals on a broader pilot for early-phase studies.

In a document published on April 29, the agency specifically notes that the pilot will build on NIST's principles for trustworthy AI systems: validity, safety, security, accountability, explainability, privacy protection, and fairness. Comments on the program are open until May 29, 2026.

Where it starts

In the first stage, the FDA is working with two oncology studies. AstraZeneca is already running TRAVERSE, a multicenter Phase 2 study in patients with previously untreated mantle cell lymphoma, with participation from MD Anderson and the University of Pennsylvania. Amgen is launching STREAM-SCLC, a Phase 1b study in patients with limited-stage small cell lung cancer.

For the AstraZeneca project, the agency has already confirmed it can technically receive and validate signals in real time through the Paradigm Health platform. That became the main proof that the scheme works anywhere outside a slide deck.

• Patient selection for trials
• Dose selection and escalation
• Safety signal monitoring
• Adaptive protocol design
• Earlier go/no-go decisions

The next step is a summer pilot with a wider range of sponsors. The FDA wants to learn where AI actually pays off: in first-in-human studies, oncology, rare diseases, or tasks like recruitment and biomarker selection.

The agency is also gathering input on how to compare the AI-based approach with the traditional process, which metrics should count as success, and what infrastructure needs to be built so companies, research centers, and the regulator can exchange data securely.

Benefits and risks

The potential gain isn't just that drugs reach the next phase faster. If the regulator sees critical signals right away, teams can adjust dosing earlier, assess safety more precisely, and decide faster whether to continue or stop a program.

According to FDA leadership, the new model should challenge the long-held assumption that a drug's path to market inevitably stretches over 10–12 years. For companies, it's also a chance to eliminate some of the administrative "dead time" that adds no scientific value but slows development.

"For sixty years we've run clinical trials in almost the same way, and

important signals could take years to reach the FDA."

But as speed goes up, so do the demands on system quality. The regulator is asking the industry directly how to measure model accuracy, track drift, verify that results hold up across different sites, and keep control over data privacy and integrity.

Explainability is a separate issue. If AI recommends speeding up the move between phases or helps screen out patients, everyone involved needs to understand why the system reached that conclusion. So for now, this is not about automated drug approval. It's about earlier, denser, and more measurable analytics.

What it means

If the pilot meets expectations, the FDA will start to change its own role: from a regulator that sees a study after the fact to a participant in more continuous monitoring.

For pharma, it's a chance to cut months from the bureaucratic cycle without formally loosening requirements. For patients, it's a chance to get to promising therapies faster, especially in oncology and rare diseases.

But the main test is still ahead: can AI genuinely speed up clinical science without turning data quality and safety into collateral damage?