Independent and not affiliated with the FDA, MHRA, ISPE, PDA, or any agency. Get the appgoutham@madhadi.com
madhadi.comData Integrity & GxP Quality
Browse all topics -> Articles Templates & Procedures Learning paths GlossaryScenariosToolsRegulatory ReferencesLearning PathsTopics About Start here
Log Plug-and-play starting point AI & Automation

Log: AI Vendor Model Change and Re-Verification

A plug-and-play log that tracks every vendor AI model update, prompt or retrieval-configuration change, and new data source affecting regulatory or CMC content production, the re-verification performed, its outcome, and the reliance decision, with field definitions and a filled specimen.

Document type: Log

Read and copy the template below into your own quality system. It is a generic starting point for your own internal use, provided as is, with no warranty; see the Terms and License. Adopting it does not by itself create compliance.

This is a ready-to-use log for tracking changes to an AI capability used in regulatory and CMC content production: a vendor’s base-model update, a change to the drafting prompt or retrieval configuration, or a new source added to the model’s grounding. Each of these is a change you may not have initiated, and each requires a defined re-verification before continued reliance. Replace every <<FILL: ...>> placeholder, keep one row per change event, and maintain the log under document control. A field-definition table, the blank log, and a filled specimen follow. Verify each cited regulation against the current source.

Field definitions

FieldDefinition
Change IDUnique identifier for this change event
Date detected / announcedWhen the change was noticed or the vendor announced it
Change typeVendor base-model update / Prompt or retrieval-config change / New source data added / Guardrail change
System / capability affectedThe named AI capability and version prior to the change
Detected byHow the change was found: vendor notice, output-quality drift noticed by a reviewer, scheduled check
Re-verification performedThe known-content check or representative re-test run against the changed capability
Re-verification outcomePass / Fail / Pass with new limitations
Reliance decisionContinue reliance / Hold reliance pending fix / Reliance resumed with added controls
Reviewed byNamed reviewer and date

Log

Change IDDate detectedChange typeSystem / capabilityDetected byRe-verification performedOutcomeReliance decisionReviewed by
<<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>>
<<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>><<FILL>>

Instructions

  1. Open a new row the same day a vendor model change is announced or detected, treating it as an uninitiated change per the governing SOP, before any further reliance on the capability for filed content.
  2. Run the re-verification against a known set of representative content (the same source documents used to validate the original configuration) and record the actual outcome, not just “checked.”
  3. If the outcome shows degraded quality or a new failure pattern, hold reliance, document the added controls or the fix required, and do not resume until re-verification passes.
  4. Retain this log as a GxP record supporting the validated state of the AI capability; reference it from the periodic review of the AI-assisted process.
  5. A “Pass with new limitations” outcome must state the limitation explicitly (for example, a category of claim the model now handles less reliably) so reviewers can watch for it going forward.

Acceptance criteria

  • Every known vendor model change, prompt change, retrieval-source addition, or guardrail change affecting the capability has a row, including changes detected by output drift rather than a vendor announcement.
  • No change proceeds to continued reliance without a recorded re-verification outcome.
  • A failed or limited outcome blocks reliance until resolved, with the resolution recorded in a follow-up row or cross-reference.
  • The log is reviewed at the periodic review of the AI-assisted regulatory content process.

Filled specimen

Change IDDate detectedChange typeSystem / capabilityDetected byRe-verification performedOutcomeReliance decisionReviewed by
AIC-2026-01103-Mar-2026Vendor base-model updateRegulatory drafting assistant v4.2 to v4.3Vendor release noticeRe-ran 12 representative study-report summarization tasks from the original validation set, compared output to prior baselinePass, no material difference in accuracy or citation behaviorContinue relianceJ. Alvarez, 05-Mar-2026
AIC-2026-01422-Jul-2026Prompt changeRegulatory drafting assistant v4.3Planned change by Data Science to improve section-heading formattingRe-ran the same 12-task setPass with new limitation: model now occasionally omits the source citation on formatting-only edits; reviewers alerted to check citation presence specifically after a formatting passReliance resumed with added reviewer checkJ. Alvarez, 24-Jul-2026

The second row shows why “Pass with new limitations” exists as its own outcome: the change did not fail outright, but it introduced a specific new failure mode that reviewers needed to know to watch for, which a simple pass/fail field would have hidden.

Common inspection findings this log prevents

  • A vendor model update taking effect with no record that anyone noticed, checked, or approved continued reliance.
  • Re-verification claimed in an SOP’s general language but no actual per-event record showing it happened.
  • A degraded output pattern discovered by a reviewer informally, with no record of the change that caused it or the decision to continue relying on the capability anyway.
  • No mechanism connecting this log to the periodic review of the AI-assisted process, so a pattern of frequent, poorly-managed vendor changes goes unnoticed.

References

Section 6 (change control) of the governing SOP for AI-assisted regulatory and CMC content. ICH Q9(R1), Quality Risk Management. Software supplier assessment record for the AI vendor, per software supplier assessment.

Confirm the current version of any cited procedure before issue.

Use madhadi.com as an app Full screen, works offline, one tap from your home screen.