FDA says Agentic AI is used for premarket reviews, inspections, and more

FDA’s brief announcement specifies the agency will use “Agentic AI” – which are AI systems “designed to achieve specific goals by planning, reasoning, and executing multi-step actions” – for more ”complex tasks” than previous uses of AI within the agency, such as:

• premarket review
• review validation
• post-market surveillance
• meeting management
• inspections and compliance and administrative functions.

FDA emphasizes in its press release that the tool is “entirely optional for FDA staff and is used voluntarily,” and entails human oversight.

This announcement comes six months after the agency’s rollout of “Elsa,” FDA’s generative AI tool designed to enhance the efficiency of its regulatory operations, as we analyzed in June online here. At that time, we weighed questions about Elsa’s intended uses, capabilities, and the existence of guardrails. The announcements for both Elsa and Agentic AI specify that Elsa is not being trained on data submitted by industry, but concerns continue to surround the government’s use of AI for product review.

FDA now claims that Elsa has been used by more than 70% of staff, with the Center for Devices and Radiological Health (CDRH) and the Center for Biologics Evaluation and Research (CBER) showing the highest adoption rate. FDA simultaneously announced a two-month “Agentic AI Challenge” for staff to build Agentic AI solutions and demonstrate them at FDA’s Scientific Computing Day in January 2026. In addition to these efforts to expedite medical product reviews, FDA said last month it was kick-starting a hiring spree to add more drug reviewers to its employee roster.

FDA has long relied on data analytics to guide its enforcement priorities, with its Office of Data, Analytics, and Research (ODAR) managing the agency’s enterprise data strategy and fostering a culture of data-driven decision-making across regulatory activities. While recent initiatives – such as the adoption of Agentic AI and generative tools like Elsa – have brought new visibility and branding to these efforts, the underlying strategy of leveraging data to target inspections, prioritize risk, and inform regulatory actions is not new. FDA’s use of AI tools like Elsa is an evolution of longstanding practices, including earlier systems such as PREDICT for import screening and risk assessment.

FDA’s Technology Modernization Action Plan (TMAP) and Data Modernization Action Plan (DMAP) further outline a multi-year strategy for improving data stewardship, analytics, and real-time use of data to accelerate enforcement and regulatory actions, emphasizing that data-driven enforcement is foundational to FDA’s mission.

However, the use of Agentic AI in premarket review is new: this is a sensitive and untested frontier. The statutory authority for sponsors to rely on data they do not own that they do not have a right of reference is limited, and there are limits to FDA’s authority to access and allow such data to inform its review of a subsequent marketing application. While the announcement states that the AI will not be trained on data submitted by industry, it remains unclear whether the AI be required to “forget” this data when it moves on to another sponsor’s application.

HHS AI Strategy

This move comes within the broader context of the Trump administration directing federal agencies to take an AI-first approach. Most recently, on December 4, the Health and Human Services (HHS) Department announced a “OneHHS” AI approach to integrate machine learning (ML) technologies into the health care system through coordination across federal health agencies’ workforce, internal operations, and research.

The initiative asks FDA and other agencies “to collaborate in the development of one robust Department-wide AI infrastructure, streamlining workflows and enhancing cybersecurity.” For example, pillar four of the AI Strategy specifically directs HHS to “facilitate reproducible pipelines (pre registration, protocol standardization, data standardization, code and/or data release when feasible) to expedite validation.”

As HHS and FDA further embrace artificial intelligence, we will continue to monitor these developments closely for information on how AI is being used, especially in the application review process.

Authored by Jodi Scott, Jason Conaty, and Bryan Walsh