Regulators leaning toward digital twins

Drug regulators are starting to ask for something drug companies did not have to show a few years ago: not just the data, but a reproducible model of how the data was produced and how the process behaves under stress. In January 2025, the Food and Drug Administration published draft guidance for artificial intelligence used in regulatory decisions, built around a risk-based “credibility assessment framework” for a specific context of use. (fda.gov) A digital twin is a software copy of a real thing that gets updated with real measurements, like a flight simulator that keeps learning from an actual airplane. In drug development, that “thing” can be a patient, a manufacturing line, or a trial control group built from prior clinical and real-world data. (appliedclinicaltrialsonline.com) The Food and Drug Administration is not treating this as a fringe idea anymore. Its drug center says it has seen a significant increase in submissions with artificial intelligence components, spanning nonclinical work, clinical development, postmarketing, and manufacturing, with more than 500 such submissions counted from 2016 to 2023. (fda.gov) That shift is happening because trials have become harder to run and easier to fail. Applied Clinical Trials cited a July 2024 analysis of 66,935 studies showing that 32% of trials in the first half of 2024 were terminated during Phase II, a 56% increase over pre-pandemic levels. (appliedclinicaltrialsonline.com) One place digital twins show up first is the control arm, which is the group used to compare whether a drug actually works. Instead of enrolling as many patients into placebo or standard-care groups, sponsors can use historical and real-world data to simulate what a closely matched patient’s path would likely look like. (appliedclinicaltrialsonline.com) Regulators are not waving this through on faith. The 2025 Food and Drug Administration draft guidance says any artificial intelligence model used to support a safety, effectiveness, or quality decision needs credibility matched to its exact job, which means the same model is not automatically acceptable for a different question or dataset. (fda.gov) The agency has also moved from one-off guidance to broader rule-setting. The Food and Drug Administration’s drug center says the 2025 draft guidance was informed by more than 800 public comments, an August 6, 2024 public workshop, and its own review experience across those 500-plus submissions. (fda.gov) Industry is reading that signal as a push toward systems that can be audited instead of demos that only look impressive in a pilot. Pharmaceutical Technology wrote on January 13, 2026 that drugmakers are shifting toward agentic artificial intelligence, digital twins, and supply-chain tools that improve efficiency, but under tighter expectations for data integrity, traceability, and secure collaboration. (pharmtech.com) That combination explains why digital twins are getting more attention than flashier artificial intelligence products. A digital twin can leave a trail of inputs, assumptions, version history, and test results, which fits a regulated industry much better than a black-box tool that cannot explain why it changed its answer. (fda.gov) (pharmtech.com) The likely near-term result is not “virtual drugs” replacing lab work or human trials. It is more narrow use: simulated control arms, manufacturing process models, and decision tools that can be validated for one defined purpose, then checked again when the purpose changes. (appliedclinicaltrialsonline.com) (fda.gov) That is why the regulatory mood matters more than any single software launch. When the Food and Drug Administration starts asking how a model was built, where its data came from, and whether it is credible for one exact use, the winners are the companies that can show a working digital replica instead of a slide deck. (fda.gov) (pharmtech.com)