ResearchResearch paperSafety & Alignment · Efficiency & Inference · Large Language Models1 source · Oct 16, 2026

Multiple Turns, Short Horizons: Measuring Effective-Horizon Compression in Jailbreaks

Multi-turn jailbreaks use feedback from a victim large language model to adapt their queries over several turns.

Key points

  • We define the effective horizon as the shortest remaining query sequence that recovers the same attack progress.
  • Empirically, we find that roughly half of the original query sequence is sufficient on average to recover the same attack progress.
  • We show that this shorter history can recover the best point reached during an unsuccessful attack and provide a better starting point for continuing the attack.
  • We further find that short reward-propagation windows preserve calibrated defense performance close to full-trajectory returns.

Sources (1)

Extractive summary: sentences quoted from the sources.

Before this

  1. Oct 15, 2026openai/codex rust-v0.163.0: 0.163.0
  2. Oct 15, 2026Code, Test, and Deploy on One Machine with NVIDIA RTX Spark
  3. Oct 15, 2026137B A6.8B model from Microsoft?
  4. Oct 14, 2026anthropics/claude-code v2.1.300
  5. Oct 14, 2026Lexical Persistence Without Semantic Entrenchment in AI Agent Communities
  6. Oct 14, 2026Epic says its new vibe-coded Epic Games Launcher is faster than Steam

Related