By Timothy S. Donahue
Top Takeaways:
- FDA’s new agentic AI tools could speed premarket and compliance workflows that directly affect tobacco and nicotine product reviews.
- All AI outputs require human validation, with no training on industry submissions — limiting regulatory-risk exposure for manufacturers.
- With Elsa already used by 70% of FDA staff, the expansion signals a long-term modernization push likely to influence future enforcement and surveillance trends.
The U.S. Food and Drug Administration officially announced the widespread implementation of “agentic AI” capabilities, providing all agency staff with access to advanced multi-step artificial intelligence workflows. According to the statement, the tools are optional and designed to support a variety of regulatory tasks — from premarket reviews and post-market surveillance of nicotine products to inspections, compliance, and administrative duties.
The agency emphasized that all AI tools will operate within a secure government cloud (“GovCloud”) environment and will not be trained on data submitted by regulated industries — a safeguard the FDA states ensures sensitive regulatory and research data remain protected from external influence.
Marty Makary, the FDA commissioner, described the rollout as part of a larger modernization effort. “We are diligently expanding our use of AI to give our reviewers, scientists, and investigators the best possible tools,” he said. “There has never been a better time in agency history to modernize with tools that can significantly improve our ability to accelerate more cures and meaningful treatments.”
Jeremy Walsh, the agency’s Chief AI Officer, stated that agentic AI will provide staff with “a powerful tool to streamline their work and help them ensure the safety and efficacy of regulated products,” including those regulated by the FDA such as tobacco and nicotine products.
The agentic-AI rollout follows the earlier mid-2025 launch of a generative AI system called Elsa, which is already used by more than 70% of FDA employees. Elsa has been utilized to assist with reading, summarizing, drafting documents, reviewing clinical protocols, summarizing adverse events, and targeting inspections.
An FDA spokesperson emphasized that the new AI tools are “exploratory” and that “all outputs from AI are reviewed and validated by FDA staff before being incorporated into any official regulatory action.” The agency clarified that AI tools do not make regulatory decisions or replace human judgment.
The same report highlighted some industry and public-interest concerns that agentic AI — defined as AI capable of planning, reasoning, and executing multi-step tasks — can behave unpredictably. This makes robust oversight and human-in-the-loop review essential.
As part of the deployment, FDA launched a two-month “Agentic AI Challenge,” inviting staff to develop prototype AI workflows; selected projects will be showcased at the agency’s Scientific Computing Day in January 2026.
