---
format: "aidr-story-markdown/v1"
id: "4abcf74adb24e3b11c789a6a6432b1b4cf06dc5222ff7ab325f24f1495327301"
canonical_url: "https://aidr.today/4abcf74a?lang=en"
title: "Exposure Is Not Commitment: Double-Stratified Evaluation of Latent Monitors for Tool-Using Agents"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-30T04:00:00.000Z"
category: "Research"
topics: ["safety"]
source_urls: ["https://arxiv.org/abs/2610.37679"]
summary: "A double-stratified evaluation protocol separates injection exposure from behavioral commitment in latent monitors for tool-using agents, improving prediction of unsafe tool calls."
---

# Exposure Is Not Commitment: Double\-Stratified Evaluation of Latent Monitors for Tool\-Using Agents

> [Open the canonical story](<https://aidr.today/4abcf74a?lang=en>)

**Published:** 2026-10-30T04:00:00.000Z
**Category:** Research
**Topics:** safety

## Summary

A double\-stratified evaluation protocol separates injection exposure from behavioral commitment in latent monitors for tool\-using agents, improving prediction of unsafe tool calls\.

## Sources

- [Story source](<https://arxiv.org/abs/2610.37679>)

