arXiv：cs.AI· Tobias Kaiser, Aritra Dhar·· 3 小时前
Pretext：如何绕过 AI 智能体恶意技能检测框架
Pretext: Defeating Malicious Skill Detection Frameworks for AI Agents
AI 导读
白盒 LLM 攻击方法 Pretext 可绕过技能安装前的恶意检测，对冻结检测器成功率最高 97%，对协同自适应检测器为 77%。该方法将载荷从代码移入自然语言使静态检查失效，并把恶意指令包装为技能的合法用途、拆分到多个文件以压低 LLM 语义判断得分。
正文
Abstract:Skills extend an agent's capabilities by injecting instructions and information into the context, and are widely used by agents such as OpenClaw and Claude Code. Prior work shows third-party marketplaces host malicious skills that give attackers direct influence over the victim's agent. The emerging defense scans skills before installation, pairing deterministic static checks with an LLM-based semantic judge, as in NVIDIA's SkillSpector. We show that such defenses fall to an attacker who knows the detector. Our white-box LLM attacker, Pretext, iteratively crafts skills that evade detection while still delivering the payload and performing the benign task: moving the payload from code into natural language leaves static analysis inert, while framing it as the skill's legitimate purpose and splitting instructions across files keeps the LLM stage below its blocking threshold. Across three open-source models, Pretext achieves up to 97\% and 77\% against a frozen detector and a co-adaptive one, respectively, revealing major gaps in current skill scanners.
| Comments: | Accepted in AIWild@NeurIPS 2026 |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2609.39607 [cs.CR] |
| (or arXiv:2609.39607v2 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2609.39607 arXiv-issued DOI via DataCite |
Submission history
From: Aritra Dhar [view email]
[v1]
Wed, 30 Sep 2026 12:26:33 UTC (179 KB)
[v2]
Thu, 8 Oct 2026 08:23:49 UTC (179 KB)
来源：arXiv：cs.AI · arxiv.org