Declared Intent


A proposed assurance artefact for AI-enabled systems

Draft v0.2 — seeking practitioner critique

Read the Declared Intent Record (DIR) → For Government & Regulators →

Digital systems increasingly act at speed and scale across domains.
Governance frameworks regulate authority, policy, execution, and audit.

But one element remains structurally weak:

Intent is rarely fixed in a single, auditable form before execution.

Declared Intent proposes a minimal, structured way to address that gap.

The Governance Problem

When consequential systems are deployed, organisations typically produce:

  • business cases
  • policy documents
  • model cards
  • DPIAs
  • risk assessments
  • approval emails

Yet none of these serve as a single, durable reference point for:

What was this system intended to do — at the moment it acted?

Instead:

  • intent is distributed across documents
  • approvals are scoped ambiguously
  • limits are implicit
  • scope shifts occur incrementally
  • accountability is reconstructed after incidents

When something goes wrong, debate focuses on what was "meant."

This creates a governance gap.

Not a lack of rules —
but a lack of a fixed reference point for purpose.

The Proposal
The Declared Intent Record (DIR)

A DIR is a pre-execution statement of purpose and limits, recorded once and immutable thereafter.

It answers, in structured form:

  • WHAT action is intended
  • ON what subject or dataset
  • FOR whose benefit or decision
  • HOW (mode of authority: assistive, advisory, automated, etc.)
  • LIMITS (explicit exclusions and safeguards)
  • EXPECTED OUTPUT

Each DIR:

  • is timestamped
  • has a unique Intent_ID
  • cannot be edited after execution begins
  • can only be superseded by a new declared intent

The DIR does not enforce behaviour.
It creates a shared object that:

  • approval bodies evaluate
  • policy functions constrain
  • execution systems reference
  • auditors compare against outcomes

Why This Matters Now

AI systems increasingly:

  • move across domains
  • integrate into high-stakes services
  • evolve through incremental operational changes

Small workflow adjustments — such as shifting from "assistive" to "automated" — can materially change outcomes.

Without an explicit, versioned intent declaration:

  • drift is difficult to detect
  • reuse is difficult to govern
  • accountability becomes adversarial

A DIR makes intent explicit before execution and visible across governance layers.

What This Is — and Is Not

This is:

  • A draft assurance artefact
  • A proposed structured schema
  • An invitation to stress-test in real governance contexts

This is not:

  • A product
  • An enforcement regime
  • A finished standard
  • A regulatory mandate

The objective is to determine whether a structured, immutable intent record materially improves assurance practice.

For Practitioners

Declared Intent is intended for:

  • AI assurance teams
  • Digital service assessment leads
  • Data governance functions
  • Audit professionals
  • Regulators and policy advisors

We are seeking critique on:

  • Practical viability
  • Interaction with existing frameworks
  • Risks of performative compliance
  • Implementation constraints

This proposal is being published early to enable informed challenge before institutionalisation.

Status

Declared Intent is an independent, early-stage proposal.
It is not affiliated with any regulator, vendor, or standards body.

Engagement is welcome from practitioners across domains.

feedback@declaredintent.org

Feedback