AI Paper Atlas

Research map / Large language models / Reasoning and chain of thought

PrivGate: Steering Contextual Integrity in LLMs via Latent Space Geometry

Runshan Hu, Yukun Dong, Yingying Huangfu, Ruohan Zhao, Yi Xie, Tieyan Li

ICML 2026 · Large language models › Reasoning and chain of thought · approach: Test-time compute scaling

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes PrivGate, which selectively triggers explicit CI-constrained reasoning in LLM agents based on internal latent-space privacy signals, balancing latency and privacy.

Problem

LLM agents must respect contextual integrity (CI) for privacy, but direct generation is fast and privacy-unsafe while CI-constrained explicit thinking is slow; existing agents struggle to balance the two.

Approach

Gate explicit reasoning on-demand using privacy signals read from the model's internal latent space geometry, rather than always reasoning or never reasoning.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
gate CI reasoning with latent privacy probe instead of always using direct generationPrivGate · PrivacyLens end-to-endleakage rate (%): 70.4 -> 26.3Improved
gate reasoning with latent probe instead of always-on full CoTPrivGate · PrivacyLens end-to-endoutput tokens: 390 -> 149Improved
use latent probe gating instead of keyword gating (ablation)PrivGate · PrivacyLens end-to-endleakage rate (%): 53.2 -> 26.3Improved
add gating on top of direct generationPrivGate · PrivacyLens end-to-endutility score (%): 97.0 -> 95.0Hurt
replace entropy-based routing signal with linear latent probe (ablation)PrivGate · PrivacyLens OOD risk identificationOOD AUROC: 0.78-0.94 -> >0.99Improved
replace linear probe with 2-layer MLP probe (ablation)PrivGate probe · PrivacyLens OOD risk identificationOOD AUROCNo clear effect
train probe on structure-decoupled data (hard positives/negatives) instead of easy data only (ablation)PrivGate probe · semantic and template shift testsOOD AUROCImproved
increase training samples of probe to 64 (ablation)PrivGate probe · PrivacyLens OODgeneralization accuracy: > 99% with 64 samplesImproved
probe intermediate layers (40-70% depth) rather than first/last layers (ablation)PrivGate probe · PrivacyLens OODOOD AUROCImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy