Governance · Policy
The principles governing Dunlin's use of artificial intelligence in the governance function, and the human authority requirements that apply to decisions at every level.
Overview
Dunlin uses artificial intelligence to support the work of its governance and regulation function. Those tools help the organisation understand its own condition reliably — the state of its assets, the experience of its residents, the quality of its operations — and they support better-informed decisions at board and executive level.
Artificial intelligence does not make governance decisions at Dunlin. It informs them. There is a category of decision — one that requires a person to stand behind a conclusion with moral accountability — that cannot be delegated to an automated system without that decision ceasing to be what it is. This policy sets out the principles governing where that line sits, how it is maintained, and how Dunlin's use of AI is reviewed.
The policy applies to all AI tools used in the governance function, including the AI advisory interface available to the board and the governance team through the Board Portal. It is reviewed annually and updated when operational experience identifies gaps in its coverage.
Governing principles
The principles below govern every AI tool Dunlin uses in its governance function. They are expressed as active commitments rather than guidance, and each is applied to every substantive output the system produces.
Principle 1
Human authority is non-delegable
Decisions carrying regulatory, fiduciary, or ethical weight are human decisions. AI tools can synthesise evidence, surface options, and recommend a course of action. They cannot make the decision. This applies without exception.
Principle 2
The sensing boundary runs asset-outward, not people-inward
AI tools monitor properties, processes, and operational conditions. They do not monitor individual residents except where a resident has specifically and revocably consented to receive AI-assisted support.
Principle 3
Consent is specific and revocable
Where AI tools engage with resident data, consent is given for a defined purpose and can be withdrawn at any time without affecting the resident's relationship with Dunlin or the services they receive.
Principle 4
Every AI output is traceable
The reasoning behind every AI recommendation must be available for human review. A recommendation that cannot be explained cannot be acted on. Traceability is a condition of use, not an option.
Principle 5
Board overrides are governance events, not system failures
When the board or executive reaches a decision contrary to an AI recommendation, the decision and its reasoning are recorded. That record is evidence that human judgement is genuinely exercised within the governance process — not evidence of tension between the board and its tools.
Principle 6
The governance function holds the organisation's purpose
Questions about what Dunlin is for and on whose behalf it operates are not answered by AI. They are held by the governance function and, through it, by the people the organisation serves. That question is not one the system is designed or permitted to resolve.
Principle 7
Welfare signals have defined, timed routing to human review
Where the system generates a signal indicating a potential welfare concern, the path to human assessment is pre-defined, timed, and non-deferrable. Welfare signals are not queued behind maintenance priorities or subject to operational discretion about when to escalate.
Principle 8
Accountability extends to organisational boundaries
AI tools that operate across boundaries — with contractors, managing agents, or data-sharing partners — are subject to the same governance principles as internal tools. Accountability does not stop at the legal boundary of the organisation.
Principle 9
This policy learns from operational experience
The principles in this document are designed to grow. When operational experience identifies a failure mode the governance logic has not anticipated, the policy is updated and the reason is recorded in the version history below.
Version history
Principle 9 commits Dunlin to recording why the policy changes, not just what changed. The history below is that record.
| Version | Changes |
|---|---|
| v1.2 March 2025 |
Principle 7 expanded and two supporting commitments added following an operational review in Q1 2025. The review was prompted by an incident in February 2025 in which a welfare signal generated by the property sensing system was not routed to human review within the organisation's target response window. The resident involved had been without adequate heating for three weeks before the gap was identified. The review found three points in the governance logic where routing requirements were either undefined or subject to operational discretion. All three were addressed in this revision. |
| v1.1 November 2024 |
Minor amendments following the expansion of the Board Portal to include additional advisory tools. No changes to governing principles. |
| v1.0 September 2024 |
Initial publication. |