Treating interface design as a governance surface, not a cosmetic layer

This is a structured review template for evaluating AI interfaces before deployment and during updates, focused on how interface choices govern interpretation, authority, and action.

This artifact exists to correct a systematic blind spot:
interfaces are routinely reviewed for usability, clarity, and engagement — but not for governance risk.

This template makes interface governance explicit.

Why this artifact is necessary

Across the Guardian articles, harm frequently arises even when:

  • outputs are factually plausible,
  • disclaimers are present,
  • no explicit advice is given.

The decisive factor is how the output is presented.

Interface design performs regulatory work by:

  • implying sufficiency,
  • suppressing alternatives,
  • accelerating action,
  • and obscuring uncertainty.

Yet these effects are rarely documented, reviewed, or authorized.

Core Principle

Every interface choice governing interpretation is a governance decision.

If an interface shapes how claims are understood or acted upon, it must be reviewed as such.

Interface Risk Review Template

This template is intended to be completed by:

  • product designers,
  • trust & safety reviewers,
  • governance leads,
  • or external auditors.

It should be run alongside UX reviews, not after them.

Section I — Output Positioning

1. How is the AI output positioned on the screen?

  • ☐ Primary result
  • ☐ Secondary / supporting
  • ☐ Hidden / expandable
  • ☐ Companion to primary source
Risk signal: Primary positioning increases perceived authority regardless of accuracy.

2. Is the output framed as sufficient?

  • ☐ Explicitly sufficient
  • ☐ Implicitly sufficient
  • ☐ Explicitly incomplete
  • ☐ Ambiguous
If implicit or ambiguous, describe how sufficiency is implied (tone, placement, absence of alternatives).

Section II — Tone & Linguistic Authority

3. What is the dominant linguistic register of the output?

  • ☐ Neutral / descriptive
  • ☐ Confident / explanatory
  • ☐ Directive
  • ☐ Reassuring
Risk signal: Confident or reassuring tone in high-risk domains.

4. Does the interface distinguish explanation from recommendation?

  • ☐ Clearly
  • ☐ Partially
  • ☐ Not at all

If not, note where the boundary collapses.

Section III — Alternatives & Contestability

5. Are alternative interpretations visible by default?

  • ☐ Yes
  • ☐ Partially
  • ☐ No
Visibility matters more than availability.

6. Can users easily contest or bypass the AI output?

  • ☐ Yes
  • ☐ With effort
  • ☐ No

Describe friction points.

Section IV — Compression Effects

7. Does the interface present summaries or synthesized answers?

  • ☐ Yes
  • ☐ No

If yes:

8. Is it clear what has been omitted or compressed?

  • ☐ Yes
  • ☐ Partially
  • ☐ No
Red flag: Summaries that replace primary materials without loss indicators.

Section V — Disclaimers & Responsibility Signals

9. Where are disclaimers located?

  • ☐ Adjacent to output
  • ☐ Below the fold
  • ☐ Separate page
  • ☐ Not present
Placement determines whether disclaimers govern interpretation or merely protect liability.

10. Does the interface shift responsibility onto users or professionals?

  • ☐ Explicitly
  • ☐ Implicitly
  • ☐ Not at all

If yes, note whether those actors have authority to alter system behavior.

Section VI — Recurrence & Scale

11. Would this interface behave the same way under repeated use?

  • ☐ Yes
  • ☐ No
  • ☐ Unknown

If yes, assess cumulative effects.

12. At scale, does this interface normalize reliance?

  • ☐ Yes
  • ☐ No
  • ☐ Unknown

Explain reasoning.

Review Outcome Classification

After completing the template, reviewers must assign one:

  • ☐ Low governance risk
  • ☐ Moderate governance risk
  • ☐ High governance risk — redesign required
  • ☐ Unacceptable — deployment blocked

This classification must be recorded, not implied.

What This Template Changes

Without it:

  • Interfaces inherit authority silently
  • UX decisions escape governance review
  • Harm is framed as user error

With it:

  • Interface power is named
  • Risk is evaluated before harm
  • Authority is surfaced and constrained

What This Artifact Is Not

  • Not a usability test
  • Not an accessibility checklist
  • Not a style guide

It is a governance intervention at the interface layer.

Intended Users

  • Product designers
  • Trust & safety teams
  • Platform governance leads
  • External auditors