“There's a big gap of knowledge between infosec researchers and ML security researchers. Anthropic has a bunch of column B but not enough column A. This was discussed in some detail in the recently published Attacker Moves Second paper*. ML researchers like using Attack Success Rate (ASR) as a metric for model resistance to attack, while for infosec, any …”
The Attacker Moves Second: Stronger Adaptive Attacks Bypass Defenses Against Llm Jailbreaks and Prompt Injections
How should we evaluate the robustness of language model defenses? Current defenses against jailbreaks and prompt injections (which aim to prevent an attacker from eliciting harmful knowledge or remotely triggering malicious actions, respectively) are typically evaluated either against a static set of harmful attack strings, or against computationally weak optimization methods that were not designed with the defense in mind. We argue that this evaluation process is flawed. Instead, we should evaluate defenses against adaptive attackers who explicitly modify their attack strategy to counter a defense's design while spending considerable resources to optimize their objective. By systematically tuning and scaling general optimization techniques-gradient descent, reinforcement learning, random search, and human-guided exploration-we bypass 12 recent defenses (based on a diverse set of techniques) with attack success rate above 90% for most; importantly, the majority of defenses originally reported near-zero attack success rates. We believe that future defense work must consider stronger attacks, such as the ones we describe, in order to make reliable and convincing claims of robustness.
候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。
HN 书评
推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及- Stronger Adaptive Attacks Bypass Defenses Against LLM Jailbreaks ↗belter · 2025-10-13 · 2 分
- Stronger Adaptive Attacks Bypass Defenses Against LLM Jailbreaks ↗baxtr · 2025-11-16 · 1 分
- Attacker Moves Second: Adaptive Attacks Bypass Defenses Against LLM Jailbreaks ↗Anon84 · 2025-11-17 · 3 分