Automation transparency

AI assistance policy

Where AI may assist the newsroom and which judgments remain human-controlled.

Last reviewed: August 28, 2026

Permitted assistance

AI may help monitor approved sources, deduplicate releases, extract candidate entities and claims, compare a draft against its evidence, propose translations, and flag stale version references. These outputs remain drafts.

Human-controlled decisions

Publication, source sufficiency, benchmark validity, ranking criteria, corrections, commercial disclosures, and legal-risk judgments require human review. A model’s confidence is not evidence.

Source-locked drafting

Editorial generation is limited to an evidence pack attached to the draft. The system must not invent missing specifications, performance results, quotes, dates, or licensing terms.

Translations

Machine-assisted translations preserve citations and claim states but receive a separate review. We do not publish a language edition when the technical meaning cannot be verified.