trends.barnas.net

phenomenon · first seen 4 Oct 2026

over-refusal

A model incorrectly refusing benign requests that resemble harmful ones.

Mentions per week

0219 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Earliest sources

Latest sources

Mentioned together with

sparse autoencoderactivation steeringactivation patchingconsole jailbreakingorthogonalizationagent skillsinput filtering guardrailprompt injectionmulti-agent system security