FDA pilots real-time AI for trials

Clinical trials still run like batch processing. Sites collect data, sponsors clean it up, and regulators see it later — sometimes much later. That lag is especially painful in early-stage studies, where dose choices, safety signals, and go-or-no-go calls can change the whole program. Now the FDA is trying to collapse that delay. On April 28, the agency said it had started two real-time clinical trial proofs of concept and opened a request for information for a broader AI-enabled pilot program launching this summer. (fda.gov) ### What is the FDA actually doing? Two things at once. First, the agency has begun proof-of-concept real-time trials with AstraZeneca and Amgen. Second, it wants outside input on a pilot program for using AI-enabled tools to improve early-phase trial design, monitoring, and decision-making. The request for information was published April 29, and comments are due May 29, 2026. (fda.gov) ### Why target early-phase trials? Because this is where uncertainty is highest and the cost of slow feedback is worst. Phase 1 and Phase 2 studies often involve small patient groups, incomplete information, and repeated judgment calls about dose selection, safety, and whether a program should keep going. The FDA says those stages are a bottlene(fda.gov) regulatory standards. (federalregister.gov) ### What does “real-time” mean here? Not magic — just fewer handoffs and less waiting. In the old model, trial sites send data to the sponsor, the sponsor analyzes it, and the FDA sees a later submission. In the real-time model, endpoints and safety signals can flow to the agency as the study progresses. That gives FDA reviewers a live view of what matters most instead of a delayed snapshot. (fda.gov) ### Which trials are in the first test? AstraZeneca is running a Phase 2 multi-site trial called TRAVERSE in treatment-naïve mantle cell lymphoma, with MD Anderson and the University of Pennsylvania participating. Amgen is running a Phase 1b trial called STREAM-SCLC in limited-stage small cell lung carcinoma. For AstraZeneca’s study, the FDA say(fda.gov)hat the plumbing works. (fda.gov) ### Where does AI fit in? The agency is not saying “let the model run the trial.” The more interesting move is narrower and more practical — use AI-enabled systems to improve efficiency, support dose selection, strengthen safety monitoring, and help with earlier go-or-no-go decisions. That matches the FDA’s broader AI posture from its 2025 draft(fda.gov)ory decisions. (federalregister.gov) ### Why does provenance matter so much? Because a live system is only useful if reviewers can trust where the data came from and how the model handled it. The request says the pilot will be guided by the NIST AI Risk Management Framework. Basically, FDA wants systems that are not just fast, but inspectable — who generated the signal, what data fed it, and whether the output holds up under scrutiny. That is a very different bar from a flashy demo. (federalregister.gov) ### Is this a bigger FDA shift? Yes. The agency has already been laying groundwork. FDA says submissions with AI components have risen significantly, with more than 500 such submissions from 2016 to 2023 informing its recent work. So this pilot is not an isolated experiment. It looks more like the next step in a broader push to make AI usable inside actual regulatory workflows. (fda.gov) ### Bottom line The important change is not just “FDA likes AI.” It’s that the agency is starting to design for continuous oversight instead of delayed review. If that sticks, drug developers will need AI systems that can perform in real time and explain themselves in real time too. (fda.gov)