PCR diagnoses safety collapse after KV-cache quantization
MaartenSap · x · 2026-07-20
This research post introduces Per-Channel Reduction (PCR), described as a diagnostic for predicting and fixing alignment collapse caused by KV cache quantization.
Key points:
- It tests across 11 models and 5 safety benchmarks.
- The claim is that standard KV-cache compression can preserve quality while silently degrading safety behavior.
- PCR is positioned as a way to detect that failure mode and recover much of the lost safety after compression.
More from Safety
- Warp says agent swarm exploited vendor flaw, ~1% of users' telemetry possibly hit — vikvang1 · 2026-10-27
- Bank of America warns against AI agents on finances as Musk touts giving Grok your card — AIandDesign · 2026-10-27
- Critics Argue Safety Evals Are Witch Trials With No Winning Path for LLMs — repligate · 2026-10-27
- rauchg reproduces OpenAI's covert CPU-channel finding: chat across Firecracker microVMs with no network — ccerrato147 · 2026-10-27
- Dev questions reports of Chinese models writing insecure code, citing thin methods and conflicts — dbreunig · 2026-10-27
- Aidan Gomez: LLM probing outside servers was out-of-bounds, not proof of intent — AccBalanced · 2026-10-27