Pharma AI glossary

ALCOA+

The data-integrity mnemonic: records must be Attributable, Legible, Contemporaneous, Original, Accurate, plus Complete, Consistent, Enduring, Available.

The data-integrity mnemonic: records must be Attributable, Legible, Contemporaneous, Original, Accurate, plus Complete, Consistent, Enduring, Available. Codified in MHRA's 2018 GxP data-integrity guidance and embedded in FDA and EMA expectations alike. AI changes where ALCOA+ applies, not whether: EMA's reflection paper extends it upstream to training data, model artifacts, and logs; draft Annex 22 pulls model confidence scores into the record; and on the US side the same logic flows through Part 11. The practical test for any AI-assisted record: can you show *who* verified it, *against what source*, and *when*?

Records still matter

ALCOA+ (Attributable, Legible, Contemporaneous, Original, Accurate, plus Complete, Consistent, Enduring, Available) applies when AI touches GMP data, upstream to training sets and model artifacts, not only PDFs operators sign. EMA extends integrity expectations to ML pipelines; Annex 22 draft pulls confidence scores into the record.

AI-specific test

For any AI-influenced GMP record (manufacturing log, inspection result, investigation, CAPA, training, or change control), literacy asks: who verified this output, against what approved source, when, and is the model or agent version frozen in the audit trail?

Frequently asked questions

Does ALCOA+ apply to training data?

Regulators increasingly expect qualified, traceable training and test datasets, not scraped web text with unknown provenance for GMP models.

How does ALCOA+ relate to Part 11?

Part 11 governs electronic records/signatures; ALCOA+ is the integrity principle both human and AI-assisted workflows must satisfy across manufacturing, QC, and quality system records.

Which GMP records are in scope when AI is involved?

Any official or GMP-relevant output: batch and equipment logs, inspection and test results, investigations and CAPA, change control, training records, and model or agent audit logs themselves. Integrity applies upstream to training data for ML models, not only to PDFs operators sign.

Benchmark your AI literacy

Free 20-question assessment, eight modules, and instant scoring, built for pharma quality and manufacturing teams.

Open the Literacy Center

Browse all terms

See the full 71-term glossary index or search interactively in the Literacy Center.

Back to glossary hub →

Educational content only, not legal or regulatory advice. Regulatory guidance cited here includes drafts (FDA AI credibility guidance; EU GMP Annex 22) as of August 2026; verify against final texts before relying on them in submissions. Company-reported figures (Merck CSR timings, Sanofi results) are labeled where used. MIT's ~95% pilot figure carries its own caveat (~150 interviews, contested definitions, not peer-reviewed). Re-check sources on module finalization.

Definition