PulseAugur
EN
LIVE 08:58:03

New Attack Recovers Local LLM Outputs Via CPU Cache Traces

By PulseAugur Editorial · [1 sources] ·

Researchers have developed a novel attack that can reconstruct text generated by locally hosted Large Language Models (LLMs) by observing CPU cache activity during the detokenization process. This method, termed "Detokenization Leaks," bypasses previous attack limitations by targeting the detokenizer, a common component in LLM inference pipelines. By using Flush+Reload and Prime+Probe techniques to monitor cache behavior, the attack can recover semantically accurate outputs from various LLM deployments, including agentic systems, across different hardware and software configurations. AI

IMPACT Highlights a new class of side-channel attacks against local LLMs, potentially impacting the security of sensitive data processed by these models.

RANK_REASON Academic paper detailing a new security vulnerability in LLM detokenization. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

New Attack Recovers Local LLM Outputs Via CPU Cache Traces

How we ranked this

Signal score
15 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Academic paper detailing a new security vulnerability in LLM detokenization. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, safety
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. arXiv cs.AI TIER_1 English(EN) · Roy Weiss, Benyamin Konstantinov, Eitam Sheetrit, Tomer Simon, Yisroel Mirsky ·

    Detokenization Leaks: Reconstructing Local LLM Outputs From Cache Traces

    arXiv:2609.06674v1 Announce Type: cross Abstract: We present a new attack that reconstructs the text generated by locally hosted LLMs by observing CPU cache activity during detokenization. Unlike prior attacks that rely on deployment-specific assumptions, such as shared data memo…

RELATED ENTITIES

RELATED TOPICS