evidoria

← Back to browse

Good practice Imported

Elsa — the FDA's Agency-Wide Generative-AI Tool for Scientific Reviewers

United States of America · Silver Spring · See the United States of America profile

Evidence: Descriptive / self-reported Top 83% 40/100 · Ask Evidence Copilot about this practice

The FDA rolled out its generative-AI assistant Elsa agency-wide after a pilot review task reportedly dropped from days to six minutes, but independent reporting found the tool hallucinating citations and studies, with no clear public definition of success or accuracy guardrails.

~2 days to ~6 minutes time
Reviewer task time reduction (single case) (2025–2026 pilot)
Elsa — the FDA's Agency-Wide Generative-AI Tool for Scientific Reviewers

Details

Maturity
Scaling
Promoter
U.S. Food and Drug Administration (FDA)
Period
2025-2026
Keywords
generative AI, regulatory review, public health administration, internal operations

Context

The FDA's Office of the Commissioner uses generative AI to help scientific reviewers process consumer complaints, adverse events and applications for products under its authority faster, through an internal assistant branded 'Elsa.'

Activities

FDA Commissioner Marty Makary said he was 'blown away' by results from the agency's first AI-assisted scientific review pilot, citing a case where one reviewer's task fell from about two days to about six minutes; the FDA then set an aggressive timeline to roll Elsa out agency-wide across all FDA centers.

Results

Independent trade-press reporting on Elsa's use found accuracy problems, including the tool generating hallucinated citations and even fabricated studies, with FDA staff describing it as unreliable when queried about basic, publicly available information on approved products.

Conclusions

Because of these limitations, Elsa is explicitly excluded from formal regulatory assessments, and reporters found it unclear how the FDA defines 'success' for the tool or what guardrails prevent AI-generated errors from influencing regulatory decisions — a gap between the celebratory rollout announcement and the tool's documented reliability.

Implementation
Indicative cost
High (€500k–€5M) — Not disclosed in available sources.
Time to results
Short (< 1 year) — Pilot completed then agency set an 'aggressive' agency-wide rollout timeline across all FDA centers in 2025–2026.
Staffing & skills
FDA Office of the Commissioner, Scientific reviewers across FDA centers

Conditions for success

  • Executive-level sponsorship (Commissioner) driving adoption
  • An initial pilot used to justify an aggressive agency-wide rollout timeline

Common failure modes

  • Independent trade-press reporting found the tool generating hallucinated citations and fabricated studies
  • FDA staff described it as unreliable when queried about basic, publicly available product information
  • Elsa is explicitly excluded from formal regulatory assessments
  • Unclear how the FDA defines 'success' for the tool or what guardrails prevent AI-generated errors from influencing regulatory decisions
  • The headline time-savings claim rests on a single anecdotal case rather than a systematic study

Where it fits

Governance type
US federal agency
Scale
agency-wide (all FDA centers)
Income level
high-income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful