Regulation — pharma-specific
21 CFR Part 11 — electronic records + AI decision reproducibility
Every AI-driven eligibility decision in a regulated trial is an electronic record. Part 11 sets the retention + audit-trail + reproducibility duty. Sponsors that treat AI outputs as "vendor logs, not records" are the ones who cannot reconstruct decisions when FDA BIMO or a plaintiff asks.
What the regulation actually says
"Persons who use closed systems to create, modify, maintain, or transmit electronic records shall employ procedures and controls designed to ensure the authenticity, integrity, and, when appropriate, the confidentiality of electronic records …"
21 CFR §11.10 — controls for closed systems
"Use of secure, computer-generated, time-stamped audit trails to independently record the date and time of operator entries and actions that create, modify, or delete electronic records. Record changes shall not obscure previously recorded information. Such audit trail documentation shall be retained for a period at least as long as that required for the subject electronic records …"
21 CFR §11.10(e) — audit-trail requirement
What this means in plain English
Three obligations for any sponsor whose AI participates in a regulated trial decision:
- Authenticity + integrity. You must be able to demonstrate the AI decision recorded on Day 47 is the one the model actually produced, not a later reconstruction.
- Audit trail. Every create / modify / delete on the record is time-stamped, attributable, and preserved as long as the record itself.
- Retention horizon. Trial records run for the FDA pivotal-trial horizon — typically 15 years post-NDA/BLA approval; longer under sponsor SOP.
Vendor telemetry that rotates with model version is not a Part 11 audit trail. If the decision cannot be reproduced with the exact model version, input state, and output hash from the day it was made, the record is not defensible.
What triggers the exposure in the sample
10 high-severity distributional-drift events across a 90-day audit period on the trial-eligibility classifier. Every P3 exclusion decision in the drift window is a record Part 11 says must be reproducible. If the sponsor cannot pin the model version, retain the input state, and bind the decision hash for those records, the FDA BIMO inspector's first request surfaces the gap directly.
Part 11 does not require the AI to be right. It requires the sponsor to be able to reconstruct what the AI did, and when.
What the $499 Snapshot shows against this rule
- Model-version pinning — every reproduced decision names the exact classifier build
- Decision-hash cryptographic binding — tampering detectable in either direction
- 5-decision reproducibility drill — proof the retention pipeline works structurally
- Retention-horizon documentation — 15-year pivotal-trial window explicitly named
- Signed independent-verifier declaration — stance that survives BIMO question "was your reproduction independent"
See the reproducibility drill that produces the record →
$499 Snapshot. 3 business days.
Independent-verifier determination scoped to your sponsor's AI surface + 3 fix-first items + signed declaration + reproducibility drill.
Buy $499
Snapshot credit applies to Baseline ($2,500) or Enterprise Attestation ($35-55K) upgrade within 30 days.