VALID Trust — Britt Biocomputing
Architecture · Supplier qualification

VALID Trust

A framework for qualifying AI suppliers in regulated life sciences. It decomposes any vendor into four pillars based on what the AI actually does in the workflow — then scopes the qualification work to exactly those pillars.

Runs as the Supplier Qualification thread through every layer of the House of AI Trust
The four pillars

What the AI does decides how you qualify it

Most vendors trigger more than one pillar. A static classifier that is fixed at inference triggers only Continuous Monitoring; an autonomous trial-design agent triggers all four. The framework's job is to make explicit which pillars apply, and to scope the work accordingly — because neither a SOC 2 report nor an ISO 42001 certificate fully answers any of them for your context of use.

Pillar 01

Generative AI Validation

Applies when the AI generates

Whenever the system produces novel content — text, structured records, draft documents — qualification concentrates on where that content comes from and whether it can be trusted and reproduced.

Qualification concentrates on
Training-data provenanceOutput reproducibilityGrounding controls
Pillar 02

Agentic AI Governance

Applies when the AI acts

Whenever the system takes autonomous action across multiple steps or tools, qualification concentrates on how far that autonomy reaches and where human oversight is wired in.

Qualification concentrates on
Action scopeTool-use scaffoldingAutonomous decision boundariesHITL / HOTL design
Pillar 03

Probabilistic Acceptance

Applies when the AI varies

Whenever outputs can vary for the same input — or the supplier can't show that they don't — qualification concentrates on whether that variance is bounded, calibrated, and quantified rather than left implicit.

Qualification concentrates on
CalibrationAcceptance criteriaUncertainty quantificationInference-variance controls
Pillar 04

Continuous Monitoring

Applies when the AI can change

Whenever the model, its data, or its operating context can change after deployment, qualification concentrates on how change is detected, versioned, and governed over time.

Qualification concentrates on
Drift detectionModel versioningChange controlRevalidation triggers
The decision logic

Five steps from vendor to qualification

The pillars define what to assess. The decision logic defines the order of operations for a real vendor in front of you.

1

Map the vendor to the pillars

Determine which of the four pillars apply, based on what the system actually does. A static classifier triggers one; an LLM authoring tool triggers three; an autonomous agent triggers all four.

2

Run the pillar-specific questions

For each applicable pillar, work through the qualification questions specific to that AI modality. The questions for Generative AI Validation are not the same as those for Agentic AI Governance.

3

Locate the validation-inheritance boundary

Map the supply chain — foundation model → wrapper vendor → hyperscaler → sponsor — and mark exactly where inherited validation ends and your own qualification work must begin. Each supplier's certifications and attestations, disclosure posture, and contractual commitments determine what you can inherit.

4

Apply the cross-cutting thread

Apply the security thread across every applicable pillar and across the boundary located in step 3: AI-specific controls beyond ISO 27001. The boundary comes first because you can't secure a handoff you haven't located.

5

Scope and issue the qualification decision

Consolidate the pillar findings, thread checks, and inheritance boundary into a scoped qualification decision: qualified, qualified-with-conditions, or not qualified — with the residual obligations that fall to the sponsor named explicitly.

Step 3, in detail

The validation-inheritance boundary

The hardest question in AI supplier qualification is not whether a vendor is trustworthy — it is where someone else's validation stops being yours to rely on. VALID Trust makes that boundary explicit across the supply chain.

Validation does not flow cleanly down this chain. Each handoff is negotiated rather than standardized, and the sponsor inherits only what is explicitly documented and contractually committed — everything else falls back to the sponsor to validate. Locating that line is what turns a vendor relationship into a defensible qualification.

Existing attestations

Where SOC 2 and ISO 42001 fit

Vendors often arrive with a SOC 2 report or an ISO 42001 certificate. Both are useful evidence about the supplier that holds them. Neither is scoped to your context of use. VALID Trust reads them as inputs to the qualification, never as a substitute for it.

Pillar / threadISO 42001SOC 2Left to VALID Trust
Pillar 01Generative AI Validation Partial
Data governance for AI systems, including provenance and quality
Minimal
Processing Integrity, only if it is in scope
Output reproducibility and grounding controls for your context of use
Pillar 02Agentic AI Governance Partial
Human oversight considered in AI impact assessment
Partial
Access control over accounts and privileged actions
Action scope, autonomous decision boundaries, HITL / HOTL design
Pillar 03Probabilistic Acceptance Minimal
Risk and impact assessment, not acceptance criteria
Minimal
Processing Integrity if in scope; how it was interpreted for AI isn't disclosed
Calibration, acceptance criteria, uncertainty quantification, inference-variance controls
Pillar 04Continuous Monitoring Partial
Monitoring of AI systems in operation
Partial
Change management and monitoring, where model updates are treated as changes
Revalidation triggers tied to your context of use; model versions evaluated against acceptance criteria
ThreadSecurity Partial
AI risk controls, typically alongside ISO 27001
Baseline
Security criteria are included in every SOC 2 report
AI-specific controls beyond that baseline
1

Map to pillars

Not covered
A system description or AI system inventory is useful input. Neither assigns pillars.

2

Pillar questions

Partial
Varies by pillar, as in the table above.

3

Inheritance boundary

Where they enter
Each attestation covers only the tier that holds it. A carved-out subservice organization marks a boundary.

4

Security thread

Largest overlap
The baseline the thread builds on.

5

Qualification decision

Never covered
The decision stays with the sponsor.

Attestations can inform steps 1 through 4. They never satisfy step 5. Each is evidence about the supplier that holds it, read against that supplier's stated scope.

Coverage levels are general. What a specific report or certificate covers depends on its stated scope, the categories included, and any carved-out subservice organizations, so read the scope statement and system description, not only the opinion. SOC 2 is an attestation report, not a certification.

Where this sits in the architecture

VALID Trust is the supplier-qualification control that makes the House of AI Trust defensible — it runs through every layer as one of two cross-cutting threads, alongside Security, rather than living in a single layer. The PVL's change-control step and the Failure Mode Taxonomy both lean on it when a third-party model is in scope.

License the VALID Trust framework

Using VALID Trust in your own qualification work? Book a licensing conversation.

© 2026 Britt Biocomputing LLC. VALID Trust by Kayla Britt is licensed under CC BY 4.0

This license covers the VALID Trust framework (its four pillars and five steps) only. Proprietary methodology, templates, and tools derived from the framework are not covered by this license and are available from Britt Biocomputing LLC under a separate license. Contact us for details.