Robust Safety Monitoring of Language Models via Activation Watermarking

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Aremu, Toluwani, Ognev, Daniil, Poppi, Samuele, Lukas, Nils
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908923228848128
author Aremu, Toluwani
Ognev, Daniil
Poppi, Samuele
Lukas, Nils
author_facet Aremu, Toluwani
Ognev, Daniil
Poppi, Samuele
Lukas, Nils
contents Large language models (LLMs) can be misused to reveal sensitive information, such as weapon-making instructions or writing malware. LLM providers rely on $\emph{monitoring}$ to detect and flag unsafe behavior during inference. An open security challenge is $\emph{adaptive}$ adversaries who craft attacks that simultaneously (i) evade detection while (ii) eliciting unsafe behavior. Adaptive attackers are a major concern as LLM providers cannot patch their security mechanisms, since they are unaware of how their models are being misused. We cast $\emph{robust}$ LLM monitoring as a security game, where adversaries who know about the monitor try to extract sensitive information, while a provider must accurately detect these adversarial queries at low false positive rates. Our work (i) shows that existing LLM monitors are vulnerable to adaptive attackers and (ii) designs improved defenses through $\emph{activation watermarking}$ by carefully introducing uncertainty for the attacker during inference. We find that $\emph{activation watermarking}$ outperforms guard baselines by up to $52\%$ under adaptive attackers who know the monitoring algorithm but not the secret key.
format Preprint
id arxiv_https___arxiv_org_abs_2603_23171
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Robust Safety Monitoring of Language Models via Activation Watermarking
Aremu, Toluwani
Ognev, Daniil
Poppi, Samuele
Lukas, Nils
Cryptography and Security
Artificial Intelligence
Computers and Society
Machine Learning
Large language models (LLMs) can be misused to reveal sensitive information, such as weapon-making instructions or writing malware. LLM providers rely on $\emph{monitoring}$ to detect and flag unsafe behavior during inference. An open security challenge is $\emph{adaptive}$ adversaries who craft attacks that simultaneously (i) evade detection while (ii) eliciting unsafe behavior. Adaptive attackers are a major concern as LLM providers cannot patch their security mechanisms, since they are unaware of how their models are being misused. We cast $\emph{robust}$ LLM monitoring as a security game, where adversaries who know about the monitor try to extract sensitive information, while a provider must accurately detect these adversarial queries at low false positive rates. Our work (i) shows that existing LLM monitors are vulnerable to adaptive attackers and (ii) designs improved defenses through $\emph{activation watermarking}$ by carefully introducing uncertainty for the attacker during inference. We find that $\emph{activation watermarking}$ outperforms guard baselines by up to $52\%$ under adaptive attackers who know the monitoring algorithm but not the secret key.
title Robust Safety Monitoring of Language Models via Activation Watermarking
topic Cryptography and Security
Artificial Intelligence
Computers and Society
Machine Learning
url https://arxiv.org/abs/2603.23171