FDA Deploys Agentic AI for All Agency Employees

The U.S. Food and Drug Administration (FDA) announced the deployment of agentic AI capabilities for all agency employees. These capabilities will enable the creation of more complex AI workflows, harnessing various AI models, to assist with multi-step tasks, said the agency.

Agentic AI refers to advanced artificial intelligence systems designed to achieve specific goals by planning, reasoning and executing multi-step actions. These systems incorporate built-in guidelines — including human oversight —to ensure reliable outcomes, said FDA. The tool is optional for FDA staff and is used voluntarily.  

“We are diligently expanding our use of AI to put the best possible tools in the hands of our reviewers, scientists and investigators,” said FDA Commissioner Marty Makary. “There has never been a better moment in agency history to modernize with tools that can radically improve our ability to accelerate more cures and meaningful treatments.”

In May, the agency deployed an LLM-based tool, Elsa, voluntarily used by more than 70% of staff, according to internal agency data. Since that initial deployment, FDA program teams have driven frequent modifications to better integrate Elsa into workflows, said the agency.

The agentic AI deployment will enable FDA staff to further advance the use of AI to assist with more complex tasks, such as meeting management, pre-market reviews, review validation, post-market surveillance, inspections and compliance and administrative functions, said the agency.

As part of the agentic AI deployment, FDA is launching a two-month Agentic AI Challenge for staff to build agentic AI solutions and demonstrate them at the FDA Scientific Computing Day in January.

“FDA's talented reviewers have been creative and proactive in deploying AI capabilities — agentic AI will give them a powerful tool to streamline their work and help them ensure the safety and efficacy of regulated products,” said FDA Chief AI Officer Jeremy Walsh.  

Built within a high-security GovCloud environment, the models do not train on input data nor any data submitted by regulated industry, safeguarding sensitive research and data handled by FDA staff, said the agency.