Editorial policy for a topic where certainty is easy to fake.

Our standard is simple: a dramatic claim does not become a fact because a famous person said it, a company published it, or a headline repeated it.

1. Evidence hierarchy

We prioritize primary and synthesis sources: peer-reviewed research where available, major consensus or multi-expert reports, official standards, public evaluation results, and original framework documents. Journalism is useful for current context but should not substitute for the underlying evidence when the evidence is public.

2. Attribution

Predictions, probabilities, motives, and contested interpretations are attributed to the person or institution making them. We do not convert a personal probability estimate into a scientific forecast.

3. Safety boundaries

Some AI-risk domains are dual-use. Our biosecurity and cybersecurity coverage explains mechanisms, evidence, safeguards, and policy questions without publishing operational instructions that would materially enable harmful acts.

4. AI-assisted production

AI tools may assist research organization, drafting, code, and visual production. The publication standard is human-directed: pages must have a clear purpose, meaningful synthesis, source attribution, editorial review, and distinct value. We do not mass-publish thin pages simply to capture search queries.

5. Corrections

Substantive factual corrections should be reflected in the article and its modified date. Style edits do not justify a new “last updated” date. When uncertainty remains, the wording should say so rather than manufacture precision.

6. Search integrity

We use descriptive titles, structured data, sitemaps, internal links, and representative images to make legitimate content easier to discover. We do not cloak content, hide keywords, fabricate reviews, or produce fake expertise signals.