# Preprint finds AI memory compression can weaken safety refusals

_Published Friday, November 6, 2026 at 11:06 PM EST · AI, Science · Latest · Tier 2 — Notable_

Researchers report in an arXiv preprint that compressing the memory used during language model inference can weaken safety refusals while leaving standard quality measures largely unchanged. Across eleven instruction-tuned models and five benchmarks containing 1,894 prompts, Mistral-7B lost 15.2% of its refusals at only 1.03x perplexity, a measure of prediction error.

The researchers propose a diagnostic that identifies how compression damages safety-related features and guides mitigation. Their training-free protocol requires approximately 35 GPU-minutes and recovers up to 97% of lost alignment with under 1% latency overhead and minimal memory overhead. The study finds no universal safe bit-width: failure thresholds differ by model.

## Sources

- [cs.LG updates on arXiv.org](https://arxiv.org/abs/2606.09864)

---
Canonical: https://techandbusiness.org/newswire/bnwauMptMq1HAz_vFmbNKg
Published: 2026-11-07T04:06:43.592Z
Story chronology: 2026-11-06T05:00:00.000Z
Retrieved: 2026-11-08T06:09:31.720Z
Publisher: Tech & Business (techandbusiness.org)