Research preview. Independent expert review is pending. The worked analyses are hypothetical.
Method and corrections Interactive field model
Inspect the system around the AI system. The model is organized by the clauses of the definition: societies, the four objects, the benefits and harmful patterns governance aims at, and the conditions it races against. Apply a cross-cutting test, narrow by lifecycle stage or level, and inspect how definitions, claims, relationships and sources travel together.
Apply a cross-cutting test Whole model Power Legitimacy Distribution Adaptability
Narrow by lifecycle stage or level Lifecycle stage Any stage Research Data collection Model development Evaluation Procurement Deployment Monitoring Incident response Retirement Level Any level Community and organisational use Corporate Sectoral National Regional International Loading interactive controls. Concept links remain available. Field model · whole model
Groups follow the clauses of the definition; they are not a sequence or causal chain. Actual relationships appear in the inspector.
Jump to selected entity Societies …how societies should shape…
actor AI developer or provider actor Deploying organization actor Affected people actor Civil society and independent researchers Incentives …the incentives…
incentive Deployment pressure incentive Competitive race dynamics incentive Liability exposure incentive Procurement leverage
Institutions …institutions…
institution Regulator or public authority institution Internal safety and governance team institution Standards body institution International institution
Infrastructures …infrastructures…
infrastructure Evaluation capacity infrastructure Compute infrastructure Data, language resources, and data labor infrastructure Energy, water, and minerals infrastructure Data work and content moderation Safeguards …and safeguards around AI…
safeguard Risk management process safeguard Impact assessment safeguard Traceability safeguard Audit record safeguard Meaningful human oversight safeguard Supervision and regulatory review safeguard Pre-deployment evaluation and red teaming safeguard Incident reporting and post-deployment monitoring safeguard Complaint and appeal routes safeguard Review and sunset provisions safeguard Compute controls Benefits and harmful patterns …so that its benefits are broadly shared and harmful patterns are addressed…
principle Accountability principle Safety principle Broadly shared benefit pattern Contestable decision pattern Concentration of power pattern Systematically unreliable provenance pattern Over-reliance and deskilling pattern Sycophancy and engagement-driven agreement pattern Homogenisation and loss of authorship Choices and conditions …while they are still choices rather than conditions.
condition Lock-in condition Governance lock-in condition Reversibility window
Safeguard · Safeguards established
Pre-deployment evaluation and red teaming Testing a system against specified capabilities, populations, and failure modes before release, by people with access, expertise, and the standing to delay it.
Tested for what, on whom, by whom, and what happens if it fails?
Tests Adaptability Power
Lifecycle Evaluation
Levels Corporate National Recorded relationships Read each statement from its first entity to its second. Status describes our mapping, not certainty of cause.
Safety — operationalized by → Pre-deployment evaluation and red teaming
Mapping: asserted
'AI should be safe' is a principle; testing against specified capabilities before deployment is a safeguard with an actor and a trigger.
Inspect evidence · source links · passage review pending Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Competitive race dynamics — erodes time and incentive for → Pre-deployment evaluation and red teaming
Mapping: hypothesized
Evaluation takes time that a race penalizes, so a safeguard that depends on voluntary restraint is weakest exactly when it matters most.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Governing AI for Humanity: Final Report (opens in a new tab) Executive summary, para. xvi (p. 8)
Governance regimes are said to need global span to avert AI arms races or a race to the bottom on safety and rights, to detect and respond to incidents spanning multiple jurisdictions, and to share AI’s benefits.
Scope note: An advisory body report representing a majority consensus of its members in their personal capacity. It is not a UN member-state decision. Page locators refer to the PDF's printed page numbers.
Passage checked 2026-09-06 · un-hlab-race-to-bottom
Evaluation capacity — enables → Pre-deployment evaluation and red teaming
Mapping: asserted
Evaluation requires people, access, compute and data; where those are scarce the safeguard exists on paper only.
Inspect evidence · 1 passage · passage review pending Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Governing AI for Humanity: Final Report (opens in a new tab) Executive summary, section C 'Common benefits', paras. xxxviii–xl (p. 14)
Effective participation depends on access to talent, compute and data. Compute is described as one of the biggest barriers to entry; the report says it is unrealistic to promise access to compute that even the wealthiest countries and companies struggle to acquire, and instead seeks to put a floor under the AI divide.
Scope note: An advisory body report representing a majority consensus of its members in their personal capacity. It is not a UN member-state decision. Page locators refer to the PDF's printed page numbers.
Passage checked 2026-09-06 · un-hlab-compute-floor
Internal safety and governance team — conducts → Pre-deployment evaluation and red teaming
Mapping: asserted
NIST places responsibility for risk decisions with named roles and leadership; the team that evaluates is the institutional site of the safeguard.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Standards body — specifies repeatable requirements for → Pre-deployment evaluation and red teaming
Mapping: asserted
Standards turn 'evaluate before release' into named tests, thresholds and documentation that different organizations can apply in the same way.
Inspect evidence · source links · passage review pending Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Claims and evidence · 2 synthesis Confidence: high Contestability: low
AI risk management is an ongoing organizational practice across the system lifecycle, not a one-time approval gate.
Inspect evidence · source links · passage review pending Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
interpretation Confidence: medium Contestability: medium
A principle states a value; in this framework, a safeguard is a specified mechanism with an actor, a trigger, a record and a consequence. Specifying the mechanism lets us inspect how the value reaches a decision.
Our reading: AIGovSpec's analytical distinction, not a universal definition or evidence that mechanisms alone are sufficient. These sources connect values to organizational arrangements and challenge routes; they do not establish our four-part definition as a consensus.
Inspect evidence · 3 passages Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law (CETS No. 225) (opens in a new tab) Overview page — introduction; 'What does the Framework Convention require states to do?': fundamental principles; remedies, procedural rights and safeguards
The Council of Europe describes the Convention as the first international legally binding treaty in the field, opened for signature on 5 September 2024, covering activities within the lifecycle of AI systems. It lists principles including transparency and oversight and accountability and responsibility, and says parties must ensure documentation sufficient for affected people to challenge decisions and the use of a system, with an effective possibility to complain to competent authorities.
Scope note: The Council of Europe overview page was reviewed, not the treaty text itself. The Convention binds only parties for which it has entered into force, and is largely framed as obligations on states to adopt measures. Signature and ratification status changes; verify before relying on it. The treaty text itself (CETS 225) was opened on 19 October 2026 for the definitions comparison.
Passage checked 2026-09-06 · coe-principles-remedies