analiza aiqbench
This research introduces Adversarial Prompt Distillation (APD), a framework that transfers jailbreaking capabilities from large language models to small language models, significantly reducing computational costs and API usage. By achieving state-of-the-art attack success rates with fewer parameters, the method exposes critical scalability gaps in current LLM defenses and offers a practical testbed for advancing AI safety protocols. The work highlights the urgent need for robust defense mechanisms against lightweight, efficient adversarial attacks that can be deployed with minimal resources.
Efficient and Stealthy Jailbreak Attacks via Adversarial Prompt Distillation from LLMs to SLMs
arXiv:2506.17231v3 Announce Type: replace Abstract: Current jailbreak attacks on large language models (LLMs) predominantly rely on LLMs themselves to generate adversarial prompts, creating a critical efficiency bottleneck: each attack requires substantial computational resources and API queries, limiting scalability and practical deployment. To overcome this limitation, we propose Adversarial Prompt Distillation (APD), a novel framework that transfers jailbreaking capabilities from LLMs to small language models (SLMs) for efficient, low-resource attacks. APD integrates three key components: (1) masked adversarial knowledge pre-training via LoRA fine-tuning, (2) dynamic temperature-controlled knowledge distillation to bridge architectural gaps, and (3) reinforcement learning-based template optimization for adaptive refinement. Extensive experiments across 12 models show that APD achieves state-of-the-art attack success rates (e.g., 96.4% ASR_k on GPT-4) while dramatically improving efficiency - generating prompts 3.7x faster with 11.3x fewer parameters than teacher models. Our work establishes the first practical framework for lightweight jailbreak attacks, exposes new vulnerabilities in LLM defenses, and provides a scalable testbed for advancing AI safety research. Our code is available at: https://github.com/lxgem/Efficient_and_Stealthy_Jailbreak_Attacks_via_Adversarial_Prompt.
Abstract:Current jailbreak attacks on large language models (LLMs) predominantly rely on LLMs themselves to generate adversarial prompts, creating a critical efficiency bottleneck: each attack requires substantial computational resources and API queries, limiting scalability and practical deployment. To overcome this limitation, we propose Adversarial Prompt Distillation (APD), a novel framework that transfers jailbreaking capabilities from LLMs to small language models (SLMs) for efficient, low-resource attacks. APD integrates three key components: (1) masked adversarial knowledge pre-training via LoRA fine-tuning, (2) dynamic temperature-controlled knowledge distillation to bridge architectural gaps, and (3) reinforcement learning-based template optimization for adaptive refinement. Extensive experiments across 12 models show that APD achieves state-of-the-art attack success rates (e.g., 96.4% ASR_k on GPT-4) while dramatically improving efficiency - generating prompts 3.7x faster with 11.3x fewer parameters than teacher models. Our work establishes the first practical framework for lightweight jailbreak attacks, exposes new vulnerabilities in LLM defenses, and provides a scalable testbed for advancing AI safety research. Our code is available at: this https URL.
Computation and Language (cs.CL); Cryptography and Security (cs.CR)
arXiv:2506.17231 [cs.CL]
(or arXiv:2506.17231v3 [cs.CL] for this version)
From: Chong Zhang Mr. [view email] [v1] Mon, 26 May 2025 08:27:51 UTC (356 KB) [v2] Mon, 22 Dec 2025 13:04:30 UTC (353 KB) [v3] Mon, 8 Jun 2026 16:01:52 UTC (353 KB)
Text integral sub licența deschisă a publicației · 213 cuvinte
Subiecte conexe
Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible
LessWrong · 5 oct. 2026
Thoughts on Claude Fable's silent safeguards
LessWrong · 10 iun. 2026
Debating RSI, the US-China Gap, and Jaggedness with JS Denain of Epoch AI
Interconnects · 22 sept. 2026
Claude Fable 5 and Claude Mythos 5
Anthropic · 9 iun. 2026