KENX AI in GxP East 2026

Human-in-the-loop must be a named hard gate, not rubber-stamp oversight

4 min read

Induced from four speakers across both days of the KENX AI in GxP East 2026 conference: Bharanitharan Rajendran’s named-human gate (Bharanitharan Rajendran), Ben O’Brien’s hard-stop oversight wording (Ben O’Brien), Robert (Bob) Wherry’s “Never Trust! Always Verify!” (Bob Wherry), and Carolyn Troiano’s Expert in the Loop (Carolyn Troiano).

Passive “human oversight” is camouflage, not a control

Ben O’Brien’s warning: the phrase “utilising human oversight during operational execution” often describes passive rubber-stamping, and automation bias will fill the gap inspectors later ask about. The fix in The intended use statement is the master control lever for GxP AI is not softer language; it is a hard stop with explicit QA review and electronic sign-off, plus named human actions that leave auditable evidence. Purolea’s April 2026 CGMP warning letter is the concrete failure mode: AI-generated SOPs entered the quality system without review, and the defense was “AI did not tell us.” That answer is not a control defect in the model. It is a missing gate, and Companies own the accuracy of AI-drafted submissions regardless of the drafting tool is the accountability that remains. Human error, not AI failure, drives GxP warning letters is the citation record that made the missing gate visible.

Name the human and make the gate non-bypassable

Bharanitharan Rajendran treats enforced human-in-the-loop as a named, credentialed reviewer who retains full accountability, not a role floating on a RACI chart. His Monday plan is blunt: no AI output enters a GxP record without review and approval by a named human; configure non-bypassable QA gates; assign reviewers and approvers by name. He reframes superficial HITL as human-centric design because nobody genuinely reviews a hundred AI-generated documents when the system makes skip easy. The control-loop claim sits beside this one (In the GxP control loop, AI acts, humans decide, and the system remembers): AI acts, humans decide, the system remembers. This note owns the gate itself: who, by name, and that they cannot be routed around. Assuming humans will follow SOPs is hope, not control is what you have if the gate can be skipped.

The person at the gate must be qualified to challenge

Carolyn Troiano’s Expert in the Loop is not a checker who clicks Approve. EIL qualifications are deep operational knowledge, regulatory fluency, and critical thinking; the job is to challenge AI outputs, decide, and defend the evidence to inspectors. Audit-trailed actions are agree, disagree with justification, escalate repeat findings. In her Trust-by-Design stack the Coach (EIL) is the second line of defense after AI controls, before regulatory validation and external audit. AI cannot be accountable. The human at the gate can, which means they must be trained and authorized to say no. That is the human role Tool-driven risk assessment beats debate because owners dilute their own ratings still reserves after the tool rates.

Explicit verification is the regulatory root, not a courtesy

Bob Wherry’s motto for critical GxP AI work is “Never Trust! Always Verify!” Implicit verification that once happened while a person wrote a document vanishes when AI drafts it, so the time AI saves has to be budgeted back into a documented human verification step. Human verification of AI output must be explicit because document creation no longer carries implicit verification is that step as its own claim. The regulatory root is already old: 21 CFR 211.68(c), in force since 2008, requires a person to check that significant automated operations performed properly. HITL in GxP is not a 2026 slogan. It is a named, qualified, challenge-capable hard gate that predates generative AI and only became visible when companies started skipping it.

Graph View

Backlinks