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.

翻译：大语言模型（LLMs）可能被滥用于泄露敏感信息，例如制造武器说明或编写恶意软件。LLM提供商依赖**监控**来检测并标记推理过程中的不安全行为。一个开放的安全挑战是**自适应**攻击者，他们精心设计的攻击能够同时实现（i）逃避检测并（ii）诱导不安全行为。由于LLM提供商无法修补其安全机制（因为他们不清楚模型如何被滥用），自适应攻击者成为主要担忧。本文将**鲁棒**LLM监控建模为安全博弈，其中知晓监控器的攻击者试图提取敏感信息，而提供商则需在低误报率下准确检测这些对抗性查询。我们的工作（i）表明现有LLM监控器易受自适应攻击者攻击，并（ii）通过**激活水印**——在推理过程中审慎引入攻击者的不确定性——设计了改进的防御策略。我们发现，在知晓监控算法但不知密钥的自适应攻击者面前，**激活水印**比防护基线方法性能提升高达$52\%$。