Research preview. Independent expert review is pending. The worked analyses are hypothetical. Method and corrections
AIAIGovSpec

Practice · eleven questions

A diagnostic for any AI governance proposal.

Apply the framework to a law, corporate policy, treaty, national strategy, technical standard, or proposed intervention. The questions help you examine its theory of change, institutional assumptions, distributional effects, operational mechanism, and capacity for revision.

11 questions · 6 worked analyses

From a proposal to questions you can investigate

  1. Name the decision, who it affects, and the instrument version or date you are examining.
  2. Answer each question with evidence where available. Write “unknown” when authority, capacity, incentives or consequences are unspecified.
  3. Use those unknowns to identify what to research or ask the responsible people. Copy your summary before leaving the worksheet.

The worksheet records your analysis. It does not read an uploaded policy, check your answers or score compliance.

Analysing a decision? Treat it as a proposal: name who would decide, what they would do, and under what conditions. Printable worksheet · Use in a course

The eleven questions

  1. 01

    What outcome is being shaped?

    In plain terms: What would be different for people if this works, and what pattern is it meant to change?

    Forces the proposal to state its theory of change: which benefit it wants to spread or which harmful pattern it wants to address, at the level of patterns rather than incidents.

    Distribution

  2. 02

    What incentives currently produce or reinforce that outcome?

    In plain terms: Who gains from doing things the current way, and what pushes people toward that?

    Reveals what the existing environment makes rational, so the proposal can be judged on whether it changes incentives or merely exhorts.

    IncentivesPower

  3. 03

    Which institution has the authority to act?

    In plain terms: Who here can actually say yes, no or stop, and who has to live with the result?

    Names the site of governance: who classifies, receives reports, can stop a deployment, or hears an appeal.

    InstitutionsLegitimacy

  4. 04

    Does that institution have the capacity, independence, expertise, legitimacy, and accountability to act?

    In plain terms: Could they really act: do they have the time, skills, independence and trust, and who holds them to account?

    Separates formal existence from operational ability, and asks who holds the institution itself to account.

    InstitutionsLegitimacyPower

  5. 05

    What infrastructure does the mechanism depend on or reshape, and who controls it?

    In plain terms: What does this depend on, such as data, devices, internet access or a supplier, and who controls that?

    Surfaces dependencies on compute, data, expertise, and networks, and the double effect of infrastructure interventions on access and concentration.

    InfrastructuresPowerDistribution

  6. 06

    What safeguards translate the objective into practice?

    In plain terms: What would stop things going wrong, notice when they do, and limit the damage, and who would do that?

    Demands the operational mechanism: the named actor, the trigger, the record, the review, and the consequence, not the principle.

    SafeguardsAdaptability

  7. 07

    Where in the AI lifecycle does the intervention operate?

    In plain terms: At what stage does this act: from research and data, through building, buying and use, to checking, fixing problems and switching off?

    Locates the intervention in research, data, development, evaluation, procurement, deployment, monitoring, incident response, or retirement, and asks what it misses elsewhere.

    AdaptabilityLifecycle axis

  8. 08

    Who decided, on what authority, and who was not represented?

    In plain terms: Who made this choice, who gave them the right to, and whose voice was missing?

    Tests legitimacy directly: the mandate behind the decision and the people and places absent from it.

    InstitutionsLegitimacyDistribution

  9. 09

    Who bears the costs and risks, and who captures the benefits?

    In plain terms: Who pays or carries the risk if it fails, and who benefits if it works?

    Follows access, ownership, returns, labor, compliance costs, and exposure to error to their actual recipients.

    Distribution

  10. 10

    What happens when the mechanism or technology scales across organizations, sectors, or borders?

    In plain terms: What changes if many schools, clinics or organisations do the same thing, across sectors or across borders?

    Asks how the arrangement behaves under scale and across levels, including races to the bottom and incompatible regimes.

    AdaptabilityPowerLevels axis

  11. 11

    What becomes difficult to reverse if governance arrives too late, and how will the governance mechanism itself be revised if it turns out to be wrong?

    In plain terms: What would be hard to undo later, and how would we notice and change course if we got this wrong?

    Applies the temporal principle in both directions: the lock-in of acting late and the lock-in of acting rigidly, and whether the mechanism specifies its own review.

    AdaptabilityTemporal principle

Worked analyses

Each analysis answers all eleven questions for one decision. All are hypothetical. Three are composed from the framework memo's examples; three more test the framework at human scale: a school, a community organisation and a child-safety regulator. None describes a real organisation or jurisdiction. The human-scale cases also list what their scenarios leave unknown.

hypothetical · Corporate levelReleasing a significantly more capable frontier modelA frontier AI developer must decide whether to release a model that is significantly more capable than anything currently deployed.Read the analysis hypothetical · National levelLicensing an imported AI triage systemA health ministry in a middle-income country must decide whether to license an AI triage system that was trained and evaluated elsewhere.Read the analysis hypothetical · Sectoral levelAdopting an AI shortlisting tool under a risk-based lawAn employer operating in the European Union must decide whether to adopt a vendor's AI system to shortlist job applicants.Read the analysis hypothetical · Community and organisational use levelA school decides whether to pilot an AI tutoring toolA secondary school's leadership must decide whether to run a one-term pilot of a commercial AI tutoring tool for pupils aged 11 to 16.Read the analysis hypothetical · Community and organisational use levelA community organisation deploys a multilingual AI assistantA community organisation must decide whether to launch an AI chat assistant that answers residents' questions about public services in several languages.Read the analysis hypothetical · National levelA child-safety regulator responds to a companion chatbot teenagers useA national child-safety regulator must decide how to respond to a companion chatbot that many teenagers use. It could issue guidance, open an investigation, require design changes within its powers, or recommend new rules.Read the analysis

Worksheet · stays in your browser

Analyze a proposal

Name a law, corporate policy, treaty, national strategy, technical standard, or proposed intervention, then answer the eleven questions.

Analysing a decision instead? For example, “Should our school adopt this tool?” Treat your decision as a proposal: name who would decide, what they would do, and under what conditions.

Drafts are stored on this browser and device when browser storage is available; otherwise they last only for this session. Clearing browser data removes saved drafts. Your answers are not sent to a server. Use Copy summary to keep a separate copy.

Proposal and answers

  1. In plain terms: What would be different for people if this works, and what pattern is it meant to change?

    Forces the proposal to state its theory of change: which benefit it wants to spread or which harmful pattern it wants to address, at the level of patterns rather than incidents.

  2. In plain terms: Who gains from doing things the current way, and what pushes people toward that?

    Reveals what the existing environment makes rational, so the proposal can be judged on whether it changes incentives or merely exhorts.

  3. In plain terms: Who here can actually say yes, no or stop, and who has to live with the result?

    Names the site of governance: who classifies, receives reports, can stop a deployment, or hears an appeal.

  4. In plain terms: Could they really act: do they have the time, skills, independence and trust, and who holds them to account?

    Separates formal existence from operational ability, and asks who holds the institution itself to account.

  5. In plain terms: What does this depend on, such as data, devices, internet access or a supplier, and who controls that?

    Surfaces dependencies on compute, data, expertise, and networks, and the double effect of infrastructure interventions on access and concentration.

  6. In plain terms: What would stop things going wrong, notice when they do, and limit the damage, and who would do that?

    Demands the operational mechanism: the named actor, the trigger, the record, the review, and the consequence, not the principle.

  7. In plain terms: At what stage does this act: from research and data, through building, buying and use, to checking, fixing problems and switching off?

    Locates the intervention in research, data, development, evaluation, procurement, deployment, monitoring, incident response, or retirement, and asks what it misses elsewhere.

  8. In plain terms: Who made this choice, who gave them the right to, and whose voice was missing?

    Tests legitimacy directly: the mandate behind the decision and the people and places absent from it.

  9. In plain terms: Who pays or carries the risk if it fails, and who benefits if it works?

    Follows access, ownership, returns, labor, compliance costs, and exposure to error to their actual recipients.

  10. In plain terms: What changes if many schools, clinics or organisations do the same thing, across sectors or across borders?

    Asks how the arrangement behaves under scale and across levels, including races to the bottom and incompatible regimes.

  11. In plain terms: What would be hard to undo later, and how would we notice and change course if we got this wrong?

    Applies the temporal principle in both directions: the lock-in of acting late and the lock-in of acting rigidly, and whether the mechanism specifies its own review.

Loading the worksheet.