AI Safety & Trust and Safety
Hallucination Disclosure Policy
Purpose
Establishes transparency requirements regarding model hallucination limitations.
Scope
All generative outputs across AnswerAI, WSAI CODE, Trainer One.
Definitions
Hallucination.
Policy Statements
Known hallucination rates and limitations disclosed in model cards and in-product disclaimers. High-stakes domains display enhanced disclaimers. [FIX] Hallucination benchmark scores are now published as a standing metric in the Transparency Report Doc. 29, tracked over time by model version, not disclosed only at release.
Responsibilities
AI Safety Research Lead owns measurement methodology; Product implements UI disclosures.
Compliance Requirements
Benchmarks re-evaluated with each major release and reported quarterly.
Enforcement
Products failing to display disclaimers blocked from release.
Reporting Process
Exceptions Process
Not applicable.
Review Process
Per major release, plus quarterly metric refresh.