AI ASSURANCE · SPEQ SYNTHESIS

AI in GxP — the assurance hub

Regulated AI is a system to be assured, not a chatbot to bolt on. This hub brings together what SPEQ publishes on AI for GxP: the registry of AI system classes, the FDA/EMA principles that define quality AI, and a toolchain of deterministic assurance aids — every recommendation reproducible and explainable, because a validated assurance decision can’t rest on a model’s opinion of itself.

REGISTRY

AI System Registry →

The 9 AI system classes regulated firms deploy — copilots, agents, ML, vision, vendor AI — each framed for assurance: Context of Use, human oversight, risk tier, and the evidence it needs.

PRINCIPLES

Good AI Practice →

The 10 FDA (CDER/CBER) + EMA Guiding Principles of Good AI Practice in drug development, quoted verbatim and read through a GxP lens, with the “qualify quality AI” source shelf.

The AI-assurance toolchain

7 free, deterministic aids that walk an AI system from Context of Use to continuous monitoring. Each shows its method and is a scoping aid, not a validated assurance system.

CSA Assurance Strategy Workbench

Right-size the assurance a system needs from its process risk.

Turn per-dimension eval results into a deterministic release verdict.

OVERALL SCORE + PASS / CONDITIONAL / FAIL VERDICT AI Agent Configuration Readiness

Check whether an AI agent’s configuration is reconstructable and approvable.

RELEASE-READY / CONDITIONAL / NOT-APPROVED + MISSING ELEMENTS AI Change Impact Assessor

Decide the reassessment a change to an AI system requires.

REASSESSMENT LEVEL + EVAL-RERUN / EVIDENCE-STALE FLAGS AI Continuous Assurance Monitor

Turn monitoring signals into an assurance state and an action.

ASSURANCE STATE + BREACHES/WARNINGS + ACTION Evidence Health & Readiness Score

Score audit readiness from evidence status — missing never counts as compliant.

READINESS % + BAND + EVIDENCE GAPS Regulatory Change Impact Simulator

Trace a requirement change through a regulatory digital twin.

CONFIRMED VS POTENTIAL IMPACTED OBJECTS BY TYPE

How the assurance flow connects

Identify the system — Register it against a class in the AI System Registry, with its Context of Use.
Classify the risk — Decision consequence, model influence, data sensitivity, and change dynamics set the risk tier and the human-oversight floor.
Scope the assurance — The CSA workbench turns process risk into a verification method and the evidence it implies.
Evaluate it — Score the system against SPEQ’s benchmark; a critical-dimension failure blocks release — the model never grades itself.
Control change — A model, prompt, or tool change maps to a reassessment level; capability expansion needs human approval.
Monitor continuously — Live signals resolve to an assurance state; a safety- or policy-critical breach suspends the system.

Everything is connected — the dependency map

A regulatory requirement is never an island. This map is rendered live from SPEQ’s knowledge graph: the relationships shown are the graph’s own edges, and each object links to its canonical page — so a change here can be traced to everything it touches.

requirement · anchorEU GMP Annex 11 §12 — Security

→ is decoded by (1)

Data Integrity (ALCOA+)topic

→ is discharged by (3)

Give every user a unique, non-shared accountcontrol Provision access by authorised rolecontrol Review granted access periodicallycontrol

→ is exercised in (1)

Computerised System Validation (CSV)discipline

→ is harmonized with (3)

21 CFR 11.10(d) — Limiting system accessrequirement 21 CFR 11.10(g) — Authority checksrequirement 21 CFR 11.300 — Controls for identification codes and passwordsrequirement

→ is part of (1)

EU GMP Annex 11 — Computerised Systemsstandard

Rendered live from SPEQ’s knowledge graph — 9 connected objects across 5 relationship types. The links shown are the graph’s own relationships, not a hand-drawn diagram.

ACROSS SPEQ’S AI ESTATE

PrinciplesGood AI PracticeThe ten guiding principles FDA and EMA issued jointly in January 2026, read through a GxP lens.Standard · draftEU GMP Annex 22 — Artificial IntelligenceThe draft annex published for consultation on 7 July 2025: what it scopes in, and what it leaves out.SystemsAI systems in regulated workEach kind of AI system, its context of use, and what qualifying and monitoring it involves.StartupsBuilding an AI imaging productA founder’s journey from research result to regulated evidence, stage by stage.AdapterNVIDIA BioNeMo governance adapterHow a governed run of a pinned biology model is recorded, and what a change costs.