**FDA Launches AI Tool Elsa Amid Criticisms of Premature Rollout and Inaccuracies**

The Food and Drug Administration (FDA) has announced the agency-wide deployment of a large language model (LLM) named Elsa, aimed at assisting employees from scientific reviewers to investigators. Announced on Monday, Elsa is designed to expedite various administrative and scientific processes within the FDA, such as accelerating clinical protocol reviews and identifying high-priority inspection targets.

Elsa's touted capabilities include summarizing adverse events for safety profile assessments, performing quick label comparisons, and generating code to help develop databases for nonclinical applications. FDA Chief AI Officer Jeremy Walsh expressed enthusiasm about the rollout, stating that Elsa marks a significant milestone in integrating AI into the FDA's operations.

FDA Commissioner Marty Makary highlighted the speed and efficiency of the rollout, noting it was implemented ahead of their target date and within budget. "I set an aggressive timeline to scale AI agency-wide by June 30," Makary said. The successful and early deployment was credited to collaboration among the FDA’s in-house experts.

However, internal feedback suggests that the launch may have been premature. According to a report from NBC News, initial tests of Elsa revealed that it provided incomplete or incorrect summaries when handling questions about FDA-approved products and other public information. Some FDA staffers described the tool as "rushed" and said the promised capabilities were overstated.

Additionally, concerns were raised that Elsa should currently be confined to administrative tasks rather than scientific ones. One employee mentioned that the rapid deployment lacked comprehensive policy considerations and necessary guardrails to ensure its effective use.

Elsa is based on Anthropic's Claude language model and has been developed by consulting firm Deloitte. Since 2020, Deloitte has received $13.8 million to develop the original database of FDA documents used in Elsa's training. In April, the firm was awarded a further $14.7 million to expand the technology across the FDA.

Previously, individual centers within the FDA managed their own AI pilots. Following cost-cutting measures in May, the AI pilot initially developed by the Center for Drug Evaluation and Research (CDER), named CDER-GPT, was selected for scaling up to an FDA-wide tool and rebranded as Elsa. However, staffers from the Center for Devices and Radiological Health (CDRH) reported issues with their AI pilot, CDRH-GPT, citing bugs and connectivity problems.

While Elsa promises to enhance efficiency within the FDA, the current feedback suggests that it still requires substantial refinement to meet its full potential.