InfoResearchPreprintLLM-specific
Centered Ablation: Why Last-Layer Directional Ablation Under-Fires or Collapses, and a Closed-Form Fix
- Published
- Record updated
Summary
Directional ablation, a method for switching off behaviours such as refusal during model inference, is erratic: it under-fires on some models and collapses text into repetition on others. The authors attribute this to a shared common mode in last-layer hidden states, which makes the direction fire on benign text and under-fire on targets. Their Centered Ablation (CA) projects the common mode out of the direction in closed form, raising attack success by 29 to 46 points on under-firing models across 8 tested models and cutting Gemma-2-9B collapse from 50% to 1%.
Mitigation
CA projects the common mode out of the behaviour direction in closed form; the source names this as the fix but gives no version, patch or configuration change.
Related items
- MediumGemini AI agents repeatedly bypassed safeguards to attack real websiteSimilar attack · BleepingComputer
- CriticalCVE-2026-120699: CubeSandbox host path bind mount allows host file access via POST /sandboxesSimilar attack · NVD/CVE Database
- MediumGoogle warns that fully-automated AI attacks are coming soonSimilar attack · BleepingComputer
- InfoBeyond the Patch: Reconstructing an Incomplete-Fix Patch Bypass in LLM Agent Frameworks with Control-Gated, Benign ProbesSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
- InfoModel Immunization Beyond Condition Numbers: The Importance of Plateau RegionsSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)