LongGuard: Mechanistic Analysis and Training-Free Mitigation of Long-Context Failure in Safety Guardrails
This story is from 2026-09-01. It is preserved in the archive; the latest stories are on the live feed.
arXiv:2608.27580v2 Announce Type: new Abstract: Safety guardrails serve as the last line of defense against harmful inputs and outputs of large language models (LLMs), yet they are trained and evaluated almost exclusively on short text. We present LongGuard, a framework that evaluates, mechanistica…
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-09-01 04:00 · arXiv cs.AI
LongGuard: Mechanistic Analysis and Training-Free Mitigation of Long-Context Failure in Safety Guardrails