Editorial purpose
AI Governance World publishes evidence-led information about artificial intelligence governance. Our goal is to distinguish what has been announced, proposed, adopted, implemented, enforced, disputed or not found. We do not present inference, reputation or repetition as documentary fact.
Source hierarchy
Every source is classified before it can support a published record. A lower-tier source cannot silently substitute for the primary evidence required by a higher-tier claim.
| Tier | Source class | Editorial use |
|---|---|---|
| 1 | Legislation, courts, governments, regulators, treaty bodies, official statistics, ISO, NIST, OECD, UNESCO and comparable public authorities | Primary evidence for legal status, public policy, mandates, official statistics and Index observations. |
| 2 | Universities, peer-reviewed publications, recognized public research institutes and open scholarly repositories | Primary research evidence and independent analysis, with methods and publication status preserved. |
| 3 | Company filings, transparency reports, safety frameworks, model cards and first-party technical documentation | Evidence of an organization’s own systems, policies, disclosures and claims. |
| 4 | Professional media and other secondary reporting | Discovery and context only. Tier 4 cannot be the sole or primary source for a published factual claim or Index score. |
Publication workflow
- Discovery. Monitored official endpoints and editorial research identify a potentially relevant development.
- Capture. The system records the source URL, publisher, publication or effective date, access time and an integrity hash or immutable source snapshot where permitted.
- Extraction. Automated tools may propose structured facts, relationships, dates and evidence passages.
- Validation. Schema rules, source-tier restrictions and consistency checks reject incomplete or unsupported candidates.
- Human review. A reviewer compares each material claim with its cited evidence. An editor authorizes publication where the workflow requires it.
- Monitoring. Published records retain verification dates and can be marked stale, corrected, superseded or archived when the source or underlying facts change.
Use of artificial intelligence
AI-assisted systems support crawling, classification, translation, deduplication, extraction and quality checks. They do not have publication authority. Generated or extracted text must remain traceable to evidence, and material claims require human review before publication. We do not create a citation that does not exist or treat model confidence as factual verification.
Status, uncertainty and negative findings
Legal and policy records preserve their formal status and relevant dates. A draft is not described as law; adoption is not described as implementation; an announcement is not described as executed spending. “Not found” means the documented search protocol did not locate sufficient evidence. It is not automatically scored or described as confirmed absence.
Index governance
The Index uses the public Index methodology, including its indicator definitions, weights, scoring anchors, missing-data rules and limitations. Observations retain source and reviewer provenance. Annual results are frozen at the edition cutoff; later verified developments enter the live knowledge graph and the next edition. Material corrections to a published edition are versioned and disclosed.
Companies, people and incidents
First-party company material is reliable evidence of what the company published, not independent proof that every claim is true. Expert profiles use public professional sources and avoid unnecessary personal information. Incident records distinguish allegations, investigations, confirmed findings, disputes and resolutions, and provide the subject a reasonable route to submit primary evidence or request correction.
Independence and conflicts
Membership, course purchases, API access, institutional contracts and sponsorship do not buy favorable coverage, verification or Index treatment. Commissioned or sponsored work is labelled. Material conflicts of interest must be disclosed internally, and a conflicted reviewer must not make the final publication decision.
Corrections and challenges
Anyone may challenge a factual claim by writing to hello@aigovernanceworld.com with the page URL, the disputed statement and supporting primary evidence. We assess the evidence, record the decision and correct verified errors promptly. Substantive changes retain revision history; a correction does not erase the source record that explains why the earlier version changed.
Policy changes
AI Governance World LLC may revise this policy as the platform, evidence standards or applicable obligations change. The effective date at the top of this page identifies the current public version.