Ten deployment routes. Ten trust dimensions. Every populated cell sourced to primary documentation, with gaps shown rather than hidden.
"When the room asks 'which of these routes can we actually defend to our CISO, our DPO, and our legal team?' — this matrix gives you the answer for each one. Not as a claim. As a sourced comparison."
This matrix maps ten AI deployment routes — models from Anthropic, OpenAI, Google, and Mistral, accessed via AWS Bedrock, Azure AI Foundry, GCP Vertex AI, and direct provider APIs — across ten trust dimensions drawn from primary vendor documentation. It does not make assertions. It surfaces what the vendors themselves have committed to, in their own words. Every gap is marked. Every conditional cell is flagged. The Foundry row is amber because the evidence warrants it.
HOW THE JOURNEY UNFOLDS
01
FILTER BY YOUR CONSTRAINT
Start with what matters to your organisation: EU or US residency, training opt-out, ZDR availability, HIPAA coverage. Apply the filter. The matrix narrows to the routes that meet it.
02
OPEN THE CELLS
Click any cell. The source panel opens — the vendor's verbatim commitment, the primary document title, the URL. You are reading their words, not ours. If a vendor has not publicly documented a commitment, the cell shows a gap — not a blank, not an assumption.
03
COMPARE ANY TWO ROUTES
Select any two rows and the grid collapses to a focused side-by-side view, with every cell that differs between the two routes highlighted. Useful when the conversation has narrowed and the room needs to see exactly where two options diverge.
04
EXPORT AND FORWARD
Filter to your shortlist. Export the view as CSV. Every row carries its source URL. Forward it to your CISO, DPO, or legal counsel with the evidence already attached.
This matrix is a technical and architectural evidence tool, not legal advice. It is populated from the current Artefact 01 evidence base; cells marked as gaps or conditional need route-specific review before customer-specific reliance.
Evidence governance
Artefacts 01 and 02 render from approved canonical files. The evidence backend checks source movement and quote position, records human decisions, and promotes only approved repairs back into the public bundle.
Canonical promoted: 5/30/2026 · Public pages remain file-based, not live database-rendered.
Canonical version
v0.2
Canonical sources
67
Human decisions
87
Promoted repairs
2
Route filters
10 of 10 routes match your criteria
Data safety
Is my data safe?
Training, provider access, processor boundaries
Data location
Where is it stored?
Region, residency, routing
Inference venue
Where does inference happen?
Where the model call is served
Verifiability
Can I verify it?
Audit, logging, evidence trail
Matrix signal
Foundry-Claude is deliberately amber. Claude in Microsoft Foundry is not treated as a normal Azure-hosted route. The matrix keeps that processor asymmetry visible in the default view.
The matrix
Ten routes. Ten dimensions. One evidence surface.
Rows fade when they do not match your constraints. They stay visible because the rejected alternatives are often as useful as the shortlist.
Deployment route
Inference venuesecurity
Data use for trainingprivacy
Region · residency · routingprivacy
Logging and retentionsecurity
ZDR / retention controlsprivacy
Private networking · control planesecurity
Provider access · sub-processor chaincontractual
Auditabilitysecurity
Legal · DPA posturecontractual
Caveats and review flagscontractual
Methodology
Canonical first, automation second.
This public matrix is rendered from canonical v0.2, promoted on 5/30/2026. Monitoring and review happen behind the scenes; public claims change only through a controlled canonical promotion.
Controlled source of truth
The public page renders from an approved canonical evidence file. Backend review state supports operations, while public content remains stable until a promoted version is released.
Evidence monitoring
The evidence checker fetches official source URLs, prefers English variants, compares stored quotes with current page text, and records redirects, access failures, and quote movement for review.
Human approval
A reviewer decides whether a finding is a false positive, a source repair, a wording change, legal review, or blocked reliance. Those decisions form the audit trail.
Promotion path
Reviewed decisions produce a versioned canonical draft. Approved repairs are promoted into the next public bundle only after diff review.