In a GxP environment, the question is not whether software has AI — it is whether the evidence chain survives it. Complere ships a small set of human-in-the-loop AI assists, keeps them outside the validated boundary by design, and documents that boundary in the validation pack instead of leaving it ambiguous. This page is the position your QA and IT reviewers can hold us to.
The rules every AI capability in Complere is built against.
AI produces drafts and summaries for a qualified person. Nothing AI generates becomes a regulated record without human review and an electronic signature — the signature is the control, and it is never AI's to give.
AI assists are explicitly excluded from the CSV/CSA validation scope, and that exclusion is written into the validation pack. Your auditors see exactly where the validated system ends and the assistive tooling begins.
When an AI capability can carry validation evidence proportionate to its risk — per the same CSA logic that scopes everything else — its boundary moves. Until then, it stays an assist. Capability follows evidence, not the other way around.
Three assists, each removing drudgery from quality work — none touching the evidence chain.
Summarises long controlled documents — a 40-page SOP into a reviewable brief — so reviewers orient faster. The summary is an aid; review and approval happen against the controlled document itself.
Drafts a plain-language account of what changed between document versions, alongside the platform's field-level version compare. The authoritative diff remains the system's controlled comparison.
Drafts training assessment questions from controlled content for a training owner to review, edit, and approve. No assessment reaches a trainee without human sign-off.
The fence, spelled out — because in GxP, what the system cannot do is part of the specification.
AI cannot create, modify, or delete a controlled record, and cannot write to the audit trail as an actor. Record changes belong to authenticated humans, attributably, under ALCOA+.
Electronic signatures require human re-authentication and signing intent under 21 CFR Part 11 §11.50. There is no AI pathway to a status transition, an approval, or a sign-off.
AI session and memory data lives in the same per-tenant database as the tenant's records, under the platform's database-per-tenant isolation model — not in a shared pool.
AI-generated content is presented as AI-generated, at the point of use. A reviewer always knows whether they are reading the controlled record or an assistive draft about it.
The AI questions QA, IT security, and procurement teams ask — answered the way we would answer them in your vendor assessment.
Yes — deliberately narrowly. Complere ships three AI assists today: document summarisation, document comparison, and draft assessment generation for training. All three are human-in-the-loop by design: they produce drafts and summaries for a qualified person to review, and their output never becomes a regulated record without human review and an electronic signature.
No — by design. The AI assists sit outside the validated boundary and are explicitly excluded from the CSV/CSA validation scope, which is documented in the validation pack rather than left ambiguous. The regulated workflows they feed — document approval, training assessment, record signing — remain fully validated, with the human review step as the control point. See the validation approach.
No. AI in Complere cannot sign, approve, or modify a controlled record, cannot write to the audit trail as an actor, and cannot trigger a status transition. Signatures require human re-authentication and intent under 21 CFR Part 11 §11.50 — there is no AI pathway around that control.
Yes. AI session and memory data lives in the same per-tenant database as the rest of the tenant's records, under the platform's database-per-tenant isolation model — it is not pooled across customers. See Security & Privacy for the broader isolation posture.
Because in a GxP context the burden of proof runs the other way. An AI capability is only as valuable as the evidence boundary around it. Complere's position is that AI should remove drudgery from quality work — summarising a 40-page SOP, drafting an assessment — while every regulated decision stays attributable to a qualified human. When that boundary can be validated and defended under inspection, the scope will grow. Not before.
Related reading: Validation approach · Electronic records & signatures · Data integrity & audit trails · Trust Center
Bring your AI governance questionnaire to a demo — we will answer it against the controls on this page, not around them.
