A working paper · research preview
AI Governance From First Principles
One definition—and the questions it teaches us to ask.
The definition
AI governance is the field of inquiry and practice concerned with how societies should shape the incentives, institutions, infrastructures and safeguards around AI to share its benefits broadly, prevent and address harm, and maintain the capacity to change course as today’s choices become tomorrow’s conditions.Read it part by part →
In brief
Since 2017, governments, companies and scholars have built laws, standards and oversight bodies for AI, often before agreeing on what AI governance is.
Most of their definitions name the toolkit, and the AI system it is used on. Few name the conditions around the system, and few write timing into the definition, even when the rest of the document discusses it.
AIGovSpec proposes a new way to understand AI governance: a definition that holds those conditions, its purpose and timing in one sentence, credits the work it builds on, and turns each of its parts into a question anyone can ask of a real decision.
Three ways in
- UnderstandFollow the argumentFrom a list of instruments to the problem governance is meant to solve, and the definition that follows.The essay · ≈14 min →
- EvidenceHow the world has defined itScholars, governments, companies and global bodies, 2017 to 2026, each linked to the passage we checked.History · 38 documents →
- Use itExamine a real decisionA law, a policy, a purchase or a treaty, worked through one question at a time.Eleven questions →
What governance shapes, how to test it, and when
- I Incentives
What behavior does the system reward?
- II Institutions
Who has authority, capacity, and accountability?
- III Infrastructures
What is AI built on, and who controls it?
- IV Safeguards
What actually prevents, detects, or limits failure?
Tested for each: Power · Legitimacy · Distribution · Adaptability
Over time
Illustrative scenarios
All 6 →- CorporateReleasing 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 →
- NationalLicensing 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 →
- SectoralAdopting 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 →
Cite this
Abinsinguza, S. (2026). AI Governance From First Principles. AIGovSpec, dataset 0.7.0. https://aigovspec.com/read. CC BY 4.0.
A paper for peer review is in preparation. Where it stands · Send a correction