FDA AI Tool Failure: Agency Rushed Rollout
- The Food and Drug Administration's new artificial intelligence tool, Elsa, is facing internal skepticism regarding its capabilities.
- These staffers believe Elsa should be limited to administrative tasks rather than scientific applications.
- Elsa is built upon Anthropic's claude LLM and is being developed by Deloitte.
Internal doubts plague the FDA’s AI tool,Elsa,as agency staff question its core capabilities and the speed of its rollout. Experts at News Directory 3 report that the AI tool, developed by Deloitte, is facing internal criticism, with some FDA employees expressing concerns about inflated expectations and a lack of proper guidelines for its use. Specifically, the concerns revolve around Elsa’s ability to perform complex scientific tasks, with some suggesting it should be limited to administrative functions. The rush to implement Elsa, built on Anthropic’s Claude LLM, has raised eyebrows, especially considering the hefty contracts awarded to Deloitte for its development and expansion. Will the FDA heed these warnings and recalibrate its AI strategy? Discover what’s next as the agency navigates these challenges.
FDA’s AI Tool Faces Internal Criticism Over Capabilities
Updated June 05, 2025
The Food and Drug Administration’s new artificial intelligence tool, Elsa, is facing internal skepticism regarding its capabilities. Some FDA staff members have described the tool as “rushed,” according to Stat news,suggesting its abilities are being overstated by agency officials and the Department of Goverment Efficiency (DOGE).
These staffers believe Elsa should be limited to administrative tasks rather than scientific applications. The FDA employees also expressed concern that the agency has not established sufficient guidelines for the tool’s use.
Elsa is built upon Anthropic’s claude LLM and is being developed by Deloitte. Since 2020, Deloitte has received $13.8 million to create the original database of FDA documents used for Elsa’s training. In April, the firm secured an additional $14.7 million contract to expand the technology across the agency. The FDA maintains that Elsa operates within a secure GovCloud environment, providing a safe platform for employees to access internal documents.
Initially, individual centers within the FDA were developing their own AI pilots.Though, following cost-cutting measures in May, CDER-GPT, the AI pilot from the FDA’s Center for Drug Evaluation and Research, was chosen to be scaled up agency-wide and rebranded as Elsa.
Staffers at the Center for Devices and Radiological Health told NBC news that their AI pilot,CDRH-GPT,contains bugs,lacks internet or internal system connectivity,and struggles with document uploads and user queries.
“I’m not sure in their rush to get it out that anyone is thinking through policy and use,” the FDA employee said.
What’s next
The FDA will likely need to address the concerns raised by its staff regarding Elsa’s capabilities and implementation to ensure its effective and appropriate use across the agency.