AI Accountability Framework
Know exactly who is accountable for every AI system you run.
A named owner, steward and reviewer for every AI system — with decision rights, sign-off authority and evidence written down so you can show it to a board, a client or an auditor.
Methodology stage: Attributable Authority™
“If an AI-assisted decision went wrong tomorrow, who would answer for it?”
The problem
Accountability is usually assumed, not assigned
- AI tools are adopted by teams, but nobody owns the outcome.
- Decision rights sit in people's heads rather than on paper.
- Clients and ISO 42001 auditors ask for named accountability you can't evidence.
What you get
Deliverables
Accountability matrix
Owner, steward and reviewer named for every AI system in scope.
Decision-rights charter
Who can approve, change, pause or retire each system.
Board reporting line
How AI accountability reaches leadership, and how often.
Evidence pack
Sign-off records ready for clients, insurers and auditors.
How it works
Four steps
- 1
Map
We confirm which AI systems are in scope and who uses them.
- 2
Assign
We agree named roles and decision rights with your leadership.
- 3
Document
We write the matrix, charter and reporting line.
- 4
Embed
We hand over sign-off templates and a review schedule.
What it looks like
A sample output
| AI system | Owner | Steward | Reviewer |
|---|---|---|---|
| CV screening assistant | Head of Talent | Recruitment Ops Lead | Compliance Manager |
| Customer email drafting | Head of Client Services | Team Lead | Operations Director |
| Invoice anomaly checks | Finance Director | Finance Analyst | External Accountant |
Extract from a sample accountability matrix (fictional data)
Fixed fee
AI Accountability Framework Pricing
£1,450–£1,950
fixed fee one-off framework, sized to the number of models in scope · optional oversight £345/mo, quarterly reviews included
Best for: Businesses mixing subscription tools, fine-tuned or open-weight models and in-house builds — and anyone who needs to show a client, insurer or auditor that a named person is responsible for each of them.
Enquire about this serviceCommon questions
How is this different from the AI ownership governance review?
Ownership governance establishes what intelligence you own and what that ownership demands — regulatory, operational, economic, security and provenance obligations. The accountability framework is the next layer: it names the person answerable for each model and defines what they are responsible for. Many businesses run the ownership review first, then the accountability framework; if your position is already clear, the framework can stand alone.
We're a small company. Can one person hold several roles?
Yes, with one exception worth respecting: the AI Reviewer should not review their own work. In a ten-person business the Owner and Steward are frequently the same person, but the Reviewer is someone else — a co-founder, a senior colleague or an external adviser — because that separation is what an auditor or insurer is actually looking for.
Do you need access to our systems?
No. This is a governance assignment, not a technical build. We work from your register, model inventory and short conversations with the people using each tool, then hand you a dated record. No admin access, no production access and no new software.
Is this the kind of document an auditor or insurer will accept?
It is written to be. The output is a dated, versioned accountability record naming each model, its class, its owner, its reviewer, its permitted scope and when it was last checked — which is the shape of evidence a client questionnaire, an insurer or an ISO 42001 assessment expects. It documents your position rather than certifying it, and it is not legal advice.
What changes when we move from rented to open-weight or internal models?
The accountability gets heavier, not lighter. With a rented model you can point at a vendor for part of the answer; with weights you host or logic you wrote, there is no one else. The framework handles that by moving the model to a different class and widening what its owner is accountable for — the review cadence shortens and the evidence expectations rise.
Works well with
Not legal advice. Trestano provides governance tooling, documentation and practical guidance. Sample outputs use fictional data.