analiza aiqbench
This research introduces LOCA, a method providing local causal explanations for why specific jailbreak prompts succeed in safety-trained LLMs. By identifying minimal changes to intermediate representations that induce model refusal, the study offers granular insight into vulnerability mechanisms across various open-source models like Gemma and Llama. This work advances AI Safety by moving beyond global interpretations to pinpoint exact failure modes in autonomous systems.
Minimal, Local, Causal Explanations for Jailbreak Success in Large Language Models
arXiv:2605.00123v2 Announce Type: replace Abstract: Safety trained large language models (LLMs) can often be induced to answer harmful requests through jailbreak prompts. Because we lack a robust understanding of why LLMs are susceptible to jailbreaks, future frontier models operating more autonomously in higher-stakes settings may similarly be vulnerable to such attacks. Prior work has studied jailbreak success by examining the model's intermediate representations, identifying directions in this space that causally encode concepts like harmfulness and refusal. Then, they globally explain all jailbreak attacks as attempting to reduce or strengthen these concepts (e.g., reduce harmfulness). However, different jailbreak strategies may succeed by strengthening or suppressing different intermediate concepts, and the same jailbreak strategy may not work for different harmful request categories (e.g., violence vs. cyberattack); thus, we seek to give a local explanation -- i.e., why did this specific jailbreak succeed? To address this gap, we introduce LOCA, a method that gives Local, CAusal explanations of jailbreak success by identifying a minimal set of interpretable, intermediate representation changes that causally induce model refusal on an otherwise successful jailbreak request. We evaluate LOCA on harmful original-jailbreak pairs from a large jailbreak benchmark across Gemma, Llama, and Qwen chat models, comparing against prior methods adapted to this setting. LOCA can successfully induce refusal by making, on average, six interpretable changes; prior work routinely fails to achieve refusal even after 20 changes. LOCA is a step toward mechanistic, local explanations of jailbreak success in LLMs. Code publicly available at https://github.com/skumar-ml/loca-jailbreaks
Abstract:Safety trained large language models (LLMs) can often be induced to answer harmful requests through jailbreak prompts. Because we lack a robust understanding of why LLMs are susceptible to jailbreaks, future frontier models operating more autonomously in higher-stakes settings may similarly be vulnerable to such attacks. Prior work has studied jailbreak success by examining the model's intermediate representations, identifying directions in this space that causally encode concepts like harmfulness and refusal. Then, they globally explain all jailbreak attacks as attempting to reduce or strengthen these concepts (e.g., reduce harmfulness). However, different jailbreak strategies may succeed by strengthening or suppressing different intermediate concepts, and the same jailbreak strategy may not work for different harmful request categories (e.g., violence vs. cyberattack); thus, we seek to give a local explanation -- i.e., why did this specific jailbreak succeed? To address this gap, we introduce LOCA, a method that gives Local, CAusal explanations of jailbreak success by identifying a minimal set of interpretable, intermediate representation changes that causally induce model refusal on an otherwise successful jailbreak request. We evaluate LOCA on harmful original-jailbreak pairs from a large jailbreak benchmark across Gemma, Llama, and Qwen chat models, comparing against prior methods adapted to this setting. LOCA can successfully induce refusal by making, on average, six interpretable changes; prior work routinely fails to achieve refusal even after 20 changes. LOCA is a step toward mechanistic, local explanations of jailbreak success in LLMs. Code publicly available at this https URL
Artificial Intelligence (cs.AI)
arXiv:2605.00123 [cs.AI]
(or arXiv:2605.00123v2 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2605.00123 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 335 cuvinte
Subiecte conexe
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026
Frontier AI Risk Report (February–March 2026)
METR · 19 mai 2026
Informe de riesgos de la IA de frontera (febrero–marzo de 2026)
METR · 19 mai 2026
Frontier Risk Report (February to March 2026)
METR · 19 mai 2026