Instance · Live · 19 June 2026 · continues 01

The Missing Middle in Pharmaceutical AI Governance

A continuation of instance 01 — Validation Is Not Trust. Same body, same problem, carried one turn further: from "the playbook wasn't built for this" to "then what do I do by Monday?"

THE GUEST BODY Petri Pohjanen THE STANCE the functional-safety one

GxP AISOTIFsafety caseGAATassurance tiers

THE FOUR CONSTANTS

The guest body

Petri Pohjanen — back from instance 01; the functional-safety mind whose automotive-and-aviation assurance lineage is the ground this translation stands on.

The stance it summoned

The functional-safety one — structured assurance from automotive and aviation, translated into pharma AI vocabulary rather than imported wholesale.

The unknown on the table

The missing middle. Between "what's the risk level?" and "can we govern it?" sits the question nobody asks: how, exactly, does this system become unsafe, misleading or unverifiable under the conditions it's actually used in?

The residue

GAAT — the GxP AI Assurance Tier. Run a use case through three stacked layers — deployment class, hazard mechanism, governance capability — and a proportional tier (plus an honest "stop" category) falls out, made inspectable by a safety case. The tier is the output; the stack is the framework.

THE DISCUSSION · WHERE THE ALIVENESS LIVES

Video or podcast · coming

THE RETICLE · THE RESIDUE YOU READ IT THROUGH

Cover — The Missing Middle in Pharmaceutical AI Governance

THE ARTEFACT

"TA #11 — The Missing Middle in Pharmaceutical AI Governance"

The Trust Architecture · Edition #11 · 19 June 2026 · Nuno Valério & Petri Pohjanen

What it leaves a stranger: a three-layer diagnostic stack you can run your own use case through — deployment class → hazard mechanism → governance capability — that outputs a proportional assurance tier (GAAT), defended by a safety case that names its own fragility. It builds directly on the safety-case argument logged in instance 01.

Read the article →

Join the discussion on LinkedIn →

REFERENCES · THE GROUND IT STANDS ON

01ISO 21448 (SOTIF) — Safety of the Intended Functionality. The scenario-space discipline — known/unknown × safe/unsafe — at the core of layer 2; failure when a system works as designed in conditions it wasn't designed for.ISO · 2022
02ISO 26262 — Road vehicles · Functional safety. The automotive lineage the assurance argument is carried from — the "other route" that converged on the same place.ISO · 2018
03The safety case & Goal Structuring Notation (GSN). The structured assurance argument that integrates the three layers into something a human can walk an inspector through.SCSC Assurance Case Working Group
04ISPE GAMP 5 (2nd ed.) — A Risk-Based Approach to Compliant GxP Computerized Systems. The validation framework the "missing middle" is outgrowing — built for software the LLM no longer resembles.ISPE · 2022
05ICH Q9(R1) — Quality Risk Management. The risk-proportionate principle a tiered assurance model rests on: effort and formality scaled to the risk to the patient.ICH · 2023

← previous · 01 Validation Is Not Trust next · 02 Sandboxes for Serious Grown-Ups →