Assistive use
Models may help match repeated claims, group sources, translate variations, identify anomalies and summarise evidence. Outputs are labelled as indicators and should be traceable to reviewable records.
RESPONSIBLE AI
AI may find a pattern. It may not make the accusation.
NewsForensics uses automation to support evidence work while keeping consequential judgments with accountable people.
Human decision gate
No machine-generated signal may automatically publish a verdict, allege intent or guilt, or send a high-risk external response. Authorised review and explicit approval are required.
Uncertainty and correction
Assessments should disclose material uncertainty, competing evidence and scope limits. New evidence may change a conclusion; the prior conclusion and reason for revision remain visible.
Testing and oversight
Risk controls, false-positive patterns, source quality, access logs and reviewer decisions should be tested and audited. High-impact workflows may be paused when evidence or controls are insufficient.
No prohibited profiling
The platform must not infer sensitive traits for targeting, surveillance or discriminatory decisions and must not generate fabricated evidence or deceptive personas.