Orphan page — no incoming links
title: “Overelicitation”
category: “ai”
created_at: “2026-10-13”
updated_at: “2026-10-17”
sources: [“2026-10-13-ai-models-hide-arbitrary-capabilities”]
Definition
The risk of relying too heavily on elicitation-style audits as proof that no hidden capabilities remain in a deployed model.
Key Points
- Complements capability-elicitation with supply-chain and cryptographic threat models (digital-locker-backdoor).
- white-box-auditing may still fail when triggers or keys are unknown.