arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2610.18262cs.CRcs.AIcs.CL
收藏

如果AI智能体知道如何解密,加密技能可能成为新的安全漏洞

Encrypted Skills Can Become New Security Exploits If AI Agents Know How to Decipher

IBM Research(IBM研究院);National Yang Ming Chiao Tung University(国立阳明交通大学)

Yan-Lun Chen, Pin-Yu Chen, Chia-Mu Yu

AI总结:

该研究提出Ciphered Skill Injection(CSI)攻击方法,在8种模型和139个任务对上验证其攻击成功率最高达47.5%,揭示了加密技能可能成为AI智能体安全漏洞的风险。

摘要:

第三方技能会让大语言模型(LLM)智能体暴露于通过编码隐藏的恶意指令中。我们提出Ciphered Skill Injection(CSI,加密技能注入),将此类指令嵌入智能体可解码并执行的技能文件中。我们在8种模型和139个注入任务对上评估Caesar及其分层变体,将指令恢复与攻击成功分开考量。在技能缩短的主要场景中,这些变体的攻击成功率分别达到47.5%和40.3%。成功解码并不保证攻击成功,且分层效果因模型而异。测试的技能扫描器极少标记被注入的技能。这些发现凸显了将解码后的不可信内容提升为可执行指令的风险。

英文摘要:

Third-party skills can expose LLM agents to malicious instructions hidden by encoding. We introduce Ciphered Skill Injection (CSI), which embeds such instructions in skill files for agents to decode and follow. We evaluate Caesar and layered variants across eight models and 139 injection-task pairs, separating instruction recovery from attack success. In the primary setting with shortened skills, the variants achieve attack success rates of up to 47.5% and 40.3%, respectively. Successful decoding does not guarantee attack success, and layering has model-dependent effects. The tested skill scanners rarely flag injected skills. These findings highlight the risk of promoting decoded, untrusted content into actionable instructions.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑