Medical Device AI Audit Trails Need Lifecycle Evidence

A complaint analyst sends a de-identified event narrative to an assistant for a draft coding summary. The complaint system preserves the regulated record. Evidence for the model route must identify the analyst, source reference, de-identification class and destination.

ai audit trail medical devices work starts at that join. The evidence has to connect the source record with the model-bound request without pretending that an AI gateway owns the underlying professional decision.

TL;DR

• FDA states that the Quality Management System Regulation became effective on February 2, 2026 and incorporates ISO 13485:2016 into 21 CFR Part 820.
• Each record should bind the authenticated caller and case reference to the data class, model destination, policy version and outcome.
• The source system keeps the formal business decision and review record.
• Gateway coverage ends at authenticated HTTP model traffic routed through the enforcement point.

AI audit trail medical devices starts with the source record

The FDA Quality Management System Regulation page states that the rule became effective on February 2, 2026 and incorporates ISO 13485:2016 into 21 CFR Part 820. The Federal Register final rule explains how the amendment aligns device current good manufacturing practice requirements with the quality-management standard while retaining FDA-specific clarifications.

An AI request should link to the controlled complaint, design or change record rather than replace it. The request evidence needs the authenticated user, intended use, approved model account, content classification, policy version and response disposition. The quality system retains approval and review authority.

A reviewer should be able to place one complaint printout beside a timestamped request record and see the same protected case reference on both.

The source record and AI record should share a protected case identifier. The records serve different purposes inside the control design: the source system proves the formal action and its approval, while the AI record proves the routed request, destination and policy decision that occurred during the work.

The request record needs decision context

A useful event begins with the identity supplied by the calling application. It names the employee or delegated agent, source application and approved use case. The audit design must protect the case reference because an unguarded identifier can reveal sensitive context on its own.

The request side records the time, detected data class and resolved endpoint. Policy metadata adds the active version, outcome and reason. Response handling adds a disposition, such as permitted for review, redacted or blocked. A correlation identifier connects the two directions without forcing the audit store to keep every word of the source material.

A medical device company should never make full prompt retention the default. Retention must follow the quality purpose and the sensitivity of the underlying record.

Ai Data Protection In Medical Devices covers the underlying information risk. The audit trail answers the narrower reconstruction question: who sent what class of information to which approved route under which rule?

Review follows the named use case

A monthly count of model calls cannot show that the approved use case operated as designed. Reviewers need samples keyed to the source record. Select a case, retrieve the model event and confirm that the authenticated caller had the declared purpose. Then compare the endpoint and policy version with the approved configuration.

Run a second sample in reverse. Start with routed model events for the use case and verify that each resolves to an authorized source record. Orphan events can expose casual experimentation or a broken join. Missing events inside a known AI-assisted case can expose an unmanaged route.

A policy change also needs a traceable release time. If a destination is approved on Monday and removed on Thursday, the evidence should show which requests ran under each version. The review record then states who sampled the events, what exception was found and how it was resolved.

Ai Governance For Medical Devices places those checks inside the wider operating model.

Integrity depends on a separate write path

The application performing the work should not have sole custody of the evidence used to assess its own behavior. A separate write path can commit the route decision before the model response returns. Signed records or another tamper-evident mechanism make later changes detectable.

The stored event should remain small and purposeful. Identity, classification, endpoint, policy and outcome usually provide the durable control facts. Full prompt or response content deserves a separate retention decision because it can duplicate sensitive material and widen access. A protected hash or content reference can preserve correlation when the source system already holds the authorized copy.

Signed audit logs for AI requests explains that pattern. The control owner should also test retrieval. A record can exist somewhere in cold storage. If it cannot be produced for one case, it has little value during an examination or incident review.

Coverage stops at the routed HTTP boundary

DeepInspect's enforcement boundary covers authenticated HTTP traffic between users or agents and LLM endpoints. At that point, a policy can inspect the supplied identity and request class, evaluate the destination and record the result before forwarding.

This boundary excludes in-device inference, laboratory models and supplier-private processing that never traverses the manufacturer's managed HTTP route. Those paths need controls and evidence from the systems that own them. A coverage statement should name every included route and excluded population rather than publish one percentage that hides the gaps.

The source system also remains authoritative for the professional or regulatory decision. A gateway record can show that a request was permitted, redacted or blocked. It cannot prove that the underlying judgment was correct.

DeepInspect

DeepInspect is a stateless proxy for authenticated HTTP traffic between users or agents and LLM endpoints. It evaluates application-supplied identity, request classification, approved destination and active policy before forwarding. Each permit, redaction, reroute or block produces a signed per-decision audit record outside the calling application's write path.

For medical device, that record can connect one routed model request to the protected source reference, caller and rule in force at the time. The organization retains responsibility for the underlying decision and for routes that bypass the proxy.

Book a demo today.