Batch record review is still a top FDA finding
In FY2025, insufficient production record review under 21 CFR 211.192 was the second most-cited drug observation on FDA Form 483s, with 164 citations, or 5.8% of the total (GMP Insiders). In 2025 warning letters, “insufficient production record review and investigation (211.192, 22 citations)” was again among the leading issues (Pharmaceutical Online).
At the same time, QA teams are under pressure to release batches faster. Paper and hybrid records make that hard: reviewers read hundreds of pages to find the handful of entries that matter.
This guide covers what the regulations require, what changed for AI in GMP in 2025-2026, and how AI batch record review can speed release without weakening data integrity.
For what draft EU Annex 22 and FDA expect from AI in batch review and release, see AI for Batch Record Review and Batch Release: What It Can Do and Where Humans Must Decide.
What the regulations require
| Requirement | Source | What it means for review |
|---|---|---|
| Complete batch records | 21 CFR 211.188 | Records “shall include complete information relating to the production and control of each batch,” including in-process and lab results, yield and who performed and checked each significant step |
| QC unit review before release | 21 CFR 211.192 | All production and control records “shall be reviewed and approved by the quality control unit” before release |
| Investigation of discrepancies | 21 CFR 211.192 | ”Any unexplained discrepancy” must be “thoroughly investigated, whether or not the batch has already been distributed,” extending to other affected batches |
| Data integrity (ALCOA) | FDA data integrity Q&A guidance, Dec 2018 | Data must be “attributable, legible, contemporaneously recorded, original or a true copy, and accurate” |
| Audit trail review | FDA data integrity Q&A guidance | ”all production and control records, which includes audit trails, must be reviewed and approved by the quality unit” |
The last row is where electronic records raise the bar. ISPE’s 2026 review of audit trail practice quotes FDA: “Audit trails that capture changes to critical data must be reviewed with each record and before final approval of the record” (ISPE Pharmaceutical Engineering). Any AI-assisted review has to cover the audit trail, not just the batch record pages.
2026 trends: AI meets GMP
Europe is writing the first GMP rules for AI
The European Commission consulted from July 7 to October 7, 2025 on a revised Chapter 4, a revised Annex 11 and a new Annex 22 on artificial intelligence (European Commission). The draft Annex 22 “allows deterministic, static models for GMP critical applications,” “excludes generative AI and models with probabilistic output,” and “embeds mandatory human oversight at critical decision points” (PwC Belgium).
That position may soften. After its June-July 2026 workshop, EMA said it is “still considering the implications of the stakeholder consultation results,” and is seeking input on controls such as guardrails that could allow broader model types (EMA). As of October 2026, Annex 22 has not been finalized.
FDA is focused on credibility for a context of use
FDA’s January 2025 draft guidance on AI for drug and biological products sets out a “risk-based credibility assessment framework” for a model’s “context of use” (FDA). It is still a draft, and its scope for manufacturing uses is not spelled out on FDA’s page. Meanwhile, FDA’s voluntary Quality Management Maturity program entered its third prototype year in 2026 (Arnold & Porter), signaling continued interest in quality systems that go beyond minimum compliance.
Review by exception is the practical model
GAMP “promotes the use of review by exception” (ISPE), which MHRA defines as “A validated search tool that identifies and documents predetermined ‘abnormal’ data or actions.” EY describes review by exception shrinking a 150-page batch record review to a three-page exception report and cutting review cycle time “from days to hours” (EY).
Data integrity stays in the spotlight
In a review of 85 drug warning letters issued in 2025, “15% of all warning letters reviewed cited DI concerns” (Pharmaceutical Online). Faster review cannot come at the cost of weaker evidence.
Where AI fits in batch record review, and where it must not
The safest pattern in 2026 is AI as a reviewer’s assistant, not the release decision-maker. That matches both the draft Annex 22’s human-oversight requirement and 21 CFR 211.192, which places review and approval with the quality control unit.
| Good uses of AI | Keep with people (or deterministic, validated rules) |
|---|---|
| Flagging missing entries, signatures and dates against the master record | The batch disposition decision |
| Highlighting out-of-range values, yield outside limits and late entries | Deciding whether a discrepancy is “explained” |
| Summarizing linked deviations, investigations and change controls | Approving investigations and CAPAs |
| Grouping audit trail events for the reviewer | Final audit trail review sign-off |
| Checking SOPs and master records against GMP requirements | Changing validated parameters or limits |
Capabilities to require
- Exception lists with source references. Every flag points to the page, field or audit trail entry behind it.
- Deterministic checks for critical rules. Limits, required fields and signature checks run as validated rules; AI assists with context and summaries.
- Complete coverage. The tool must show what it checked, so reviewers know what remains for manual review.
- Deviation linkage. Link exceptions to deviations and investigations, including the § 211.192 extension to other affected batches.
- Validation package. Intended use, risk assessment and test evidence your QA team can include in computerized system validation.
- Data integrity controls. Audit trails, access control and e-signatures for the review itself.
Evaluation checklist for batch record review software
| Evaluation area | Ask the vendor | Red flag |
|---|---|---|
| Exception accuracy | Run a historical batch with known deviations; are all found? | Misses known issues or floods reviewers with noise |
| Traceability | Does each exception cite the record field or audit trail entry? | Flags with no source |
| Audit trail coverage | Does it review audit trails of critical data, per FDA expectations? | Only reviews the record pages |
| Model type | Which checks use deterministic rules vs. AI models? | Generative AI making pass/fail calls on critical data |
| Human oversight | Who approves, overrides and signs, and is it logged? | Automatic release |
| Validation | What validation documentation is provided for GxP use? | ”Validated by vendor” only |
| Regulatory watch | How will the tool adapt as Annex 22 is finalized? | No roadmap for EU GMP AI rules |
Pilot on closed batches first, where the right answer is already known, before using the tool on live release decisions.
Start with the documents behind the batch
Before you automate record review, make sure the procedures that govern it hold up. AskDona assesses SOPs, master records and investigation reports against the GMP requirements you choose, such as 21 CFR 211.188 and 211.192, and cites the passage behind every finding. Your QA team makes every decision.
Run a sample regulatory document assessment on your batch record review SOP.
FAQ
Can AI approve a batch for release? No. Under 21 CFR 211.192, the quality control unit must review and approve production and control records before release (LII). AI can prepare and prioritize the review.
What is review by exception? A validated approach that screens records and presents only predetermined abnormal data or actions to the reviewer, as defined by MHRA and promoted by GAMP (ISPE).
Is generative AI allowed in GMP-critical applications? Under the 2025 draft EU Annex 22, no: it is limited to static, deterministic models for critical uses. EMA is reconsidering after consultation, and the annex is not final (EMA).
Do audit trails need review for every batch? FDA expects audit trails that capture changes to critical data to be reviewed with each record before final approval (ISPE).
How common are 211.192 citations? In FY2025, 164 Form 483 citations, the second most frequent drug observation (GMP Insiders).
References
All pages opened and checked by October 9, 2026.
| Source | Type | Date | Used for |
|---|---|---|---|
| eCFR: 21 CFR Part 211 Subpart J | Regulation text | Current | § 211.188 |
| LII: 21 CFR 211.192 | Regulation text (mirror) | Current | Record review, discrepancy investigation |
| FDA: Data Integrity and Compliance With Drug CGMP Q&A | Regulator guidance | Dec 2018 | ALCOA, audit trail review |
| FDA: AI to support regulatory decision-making (draft) | Regulator guidance | Jan 7, 2025 | Credibility framework |
| European Commission: Chapter 4, Annex 11, Annex 22 consultation | Regulator | Jul 7 – Oct 7, 2025 | Consultation scope |
| EMA: Annex 22 multistakeholder workshop | Regulator | Jun 30 – Jul 1, 2026 | Annex 22 still under consideration |
| PwC Belgium: Annex 22 | Consultancy | Mar 31, 2026 | Draft Annex 22 scope |
| ISPE: Audit Trail Review in GxP | Industry journal | Mar/Apr 2026 | Review by exception, audit trails |
| Arnold & Porter: FDA QMM third year | Law firm | Feb 19, 2026 | QMM program |
| GMP Insiders: FY2025 FDA inspection observations | Industry analysis | Jan 5, 2026 | § 211.192 citations |
| Pharmaceutical Online: What FDA warning letters tell us | Expert column | Dec 15, 2025 | 2025 warning letter data |
| EY: Electronic batch records | Consultancy | Apr 10, 2023 | Review by exception impact |