FDA launches AI-assisted ANDA review pilot: machine learning models to screen generic drug applications for completeness and common deficiencies before human review
The FDA Office of Generic Drugs (OGD) has announced a pilot programme using validated machine learning models to perform initial completeness and deficiency screening of abbreviated new drug applications (ANDAs) for generic medicines. The AI system, developed under the FDA's Technology Modernization Action Plan, screens submitted dossiers for common refuse-to-receive (RTR) criteria, missing bioequivalence study data, and labelling discrepancies — flagging high-confidence issues for expedited human review. The pilot does not replace scientific reviewer judgement but aims to reduce median ANDA review times, which currently exceed 30 months for standard applications.
On 24 October 2026, the FDA Office of Generic Drugs (OGD) ↗ announced a pilot programme deploying machine learning (ML) models to assist ANDA reviewers by screening incoming generic drug applications before assignment to human reviewers. The ML screening layer checks for common refuse-to-receive (RTR) criteria — missing bioequivalence (BE) study reports, labelling discrepancies vs the reference listed drug (RLD), and incomplete application modules — and flags these with confidence scores to expedited human review queues. The pilot covers solid oral dosage forms (immediate-release) in the first phase.
Background: ANDA review timelines and the generic drug backlog
Abbreviated New Drug Applications (ANDAs) must demonstrate bioequivalence to the reference listed drug (RLD) — showing that the generic and brand product have similar AUC and Cmax within 80–125% acceptance limits. FDA received approximately 3,300 ANDAs in FY2025; the median review time for standard applications has exceeded 30 months historically. The Generic Drug User Fee Amendments (GDUFA) performance goals require 90% of complete standard ANDAs to receive action within 12 months, but application volume, amendment complexity, and complete response letter (CRL) cycles inflate real-world timelines. The AI pilot aims to reduce CRL cycles by catching RTR-level deficiencies at submission. See our generic vs brand drugs guide.
How the AI model works and regulatory safeguards
The OGD ML models were developed under FDA's Technology Modernization Action Plan (TMAP) ↗ and validated against a training set of ~15,000 historical ANDAs. Model outputs are non-binding and advisory — they do not make approval/rejection decisions. Each flagged deficiency includes a confidence score and the model's extracted evidence (document page, clause). Human reviewers receive the model output alongside the full application. The initiative is part of FDA's broader AI/ML in drug development programme ↗. FDA has committed to publishing the pilot's performance metrics (precision, recall, false-positive rate) in the public domain after the first 12-month evaluation period.
Implications for generic drug access and manufacturers
Generic drugs account for ~90% of US prescriptions by volume and ~22% by spend — faster ANDA review translates directly to earlier price competition on brand drugs coming off patent. The AI pilot, if successful, could reduce the ~6-month RTR cycle overhead that affects an estimated 25% of first-cycle ANDAs. For manufacturers, the key implication is that common AI-detectable deficiencies (missing in vitro dissolution data, wrong RLD citation, labelling section gaps) will be flagged faster, enabling earlier resubmission. Complete, high-quality applications benefit most. See the FDA RTR standards ↗ for the current deficiency criteria. Related: our generic vs brand guide and bioequivalence glossary entry.
FAQ
Does a Boxed Warning mean a drug is being withdrawn?
No. A Boxed Warning ↗ (also called a Black Box Warning) is the FDA's strongest label caution, but the drug remains on the market. It means the agency has determined that a risk is serious enough to require prominent, standardised labelling. The drug continues to be prescribed under appropriate clinical conditions. See the FDA's patient guide to Boxed Warnings ↗.
Where can I read the full updated label?
All current FDA-approved prescribing information is available on DailyMed ↗, maintained by the National Library of Medicine (NLM). Search by drug name to find the most recent label version. Historical label versions are also archived there.