EVIDENCE DESK
Sources & Evidence Standards
When a page makes a serious claim about AI capability, infrastructure risk, emergency preparedness or human safety, readers should be able to see what that claim rests on.
Our source hierarchy
- Tier 1 — primary and official: government departments, emergency services, regulators, standards bodies, original research papers, model-system cards and first-party incident reports.
- Tier 2 — high-quality synthesis: respected scientific institutions, universities, major security organisations and established journalism with direct sourcing.
- Tier 3 — contextual: specialist commentary and expert analysis used to explain or challenge primary material, not replace it.
- Tier 4 — anecdotal: forums, social posts and unverified claims may identify questions but should not carry a high-stakes factual conclusion.
What we do not treat as proof
A viral screenshot is not an incident report. A cinematic AI image is not documentary evidence. A benchmark score is not automatically proof of real-world autonomy. A research demonstration under controlled conditions is not the same thing as a deployed system doing the same thing unaided.
Useful official starting points
Citations inside guides
During the content passes, major factual and high-stakes claims will be supported close to the claim rather than hidden behind a generic bibliography. Scenario pages will use evidence boxes to distinguish source material from extrapolation.
Conflicts and uncertainty
When credible sources disagree, we should say so. We do not manufacture a single confident answer simply because it is more dramatic. Probability estimates for existential or takeover risk are especially uncertain and should be attributed to the people or surveys that produced them.
Last reviewed: 21 August 2026