Overview
- Elsa was deployed agency-wide on June 3, beating its June 30 deadline and coming in under budget
- The tool is designed to help FDA reviewers summarize adverse event reports, compare drug labels, generate code for nonclinical databases and identify high-risk inspection targets
- Staff feedback warns the accelerated rollout left Elsa without adequate guardrails and that some summaries and outputs have been incomplete or inaccurate
- FDA leaders plan to iterate on Elsa’s features using real-world user feedback and to expand its applications across different centers
- A separate medical-device AI, CDRH-GPT, remains in beta testing and is handicapped by bugs in document uploads and system integration