FDA AI Warning Letter Library
Four FDA AI enforcement actions (August 2024 – August 2025) analysed for pharma QA teams. Names masked. Findings documented. Remediation mapped.
WL-01 through WL-04 — Case Studies
Each case study documents what the FDA found, what was cited, and what remediation was required. No gate — maximum visibility for pharma QA professionals.
Clinical Decision Support Algorithm — Post-Hoc Acceptance Criteria
A Phase III clinical trial AI algorithm used for patient stratification was cited for defining acceptance criteria after testing was complete. The FDA found no pre-defined validation protocol and no documented rationale for the performance thresholds selected.
Retrospective validation with pre-defined criteria · Updated SOP requiring prospective acceptance criteria · Mandatory CSV Lead sign-off before testing begins
Visual Inspection AI — No Model Version Control
A visual inspection AI system used in parenteral manufacturing had undergone three model updates with no change control records. The FDA could not verify which model version was in use at specific production dates.
Retroactive SHA-256 model hashing for all versions · Change control procedure for AI model updates · Revalidation of current model version · Batch release review for affected periods
Batch Release Algorithm — No Ongoing Performance Monitoring
A batch release AI had been validated at deployment but had no ongoing monitoring programme. Significant model drift was detected during the inspection — the model had degraded materially from its validated state.
Immediate batch review · Drift monitoring programme · Performance KPIs and alert thresholds · Monthly surveillance reporting to QA management
CRO Clinical AI — Sponsor Responsibility for Third-Party Algorithm
A pharma sponsor was cited for a clinical algorithm used by their CRO — establishing that sponsors are responsible for validating any AI used in patient care decisions, regardless of whether the AI was supplied by a third party.
Sponsor-level validation review of all CRO AI · Updated contract requirements · Vendor qualification audit · Validation documentation obtained from CRO