arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2606.22698cs.CRcs.CL
收藏

对话式LLM代理的黑盒取证

Black-Box Forensics for Conversational LLM Agents

University of California, San Diego(加州大学圣地亚哥分校)

Isadora White, Yasaman Jafari, Taylor Berg-Kirkpatrick

AI总结:

研究通过非对抗对话识别对话代理的基座模型(归因)和系统提示(指纹识别),归因准确率达98%,指纹识别对未见提示的AUC达0.943。

摘要:

随着LLM驱动的诈骗泛滥,对话式LLM代理的黑盒取证为隐藏在匿名端点后的系统提供了问责途径。在无法访问模型参数或未知隐藏系统提示的情况下,识别聊天机器人端点背后的基座模型(归因),将使调查人员能够将AI驱动的诈骗追溯到提供模型的提供商。检测两个端点是否运行完全相同的系统提示(指纹识别),即使是新颖且未见过的,也能将单个诈骗链接到犯罪网络并暴露无声的API更改。我们对这两种能力进行了实证研究。我们的归因分类器通过几轮非对抗对话,以98%的准确率识别代理背后的基座模型。系统提示的归因虽然可行,但需要为每个提示重新训练大量数据;实际中的系统提示是无界且不断变化的,使得这种方法成本高昂。为了应对这种更开放的场景,我们的交叉编码器指纹识别方法在完全未见过的系统提示上实现了0.768的AUC和0.703的F1分数,而聚合每个目标代理的50次交互对话将AUC提升至0.943。因此,通过几轮普通对话,即可对具有未见系统提示的对话代理进行鲁棒准确的指纹识别。

英文摘要:

As LLM-powered scams proliferate, black-box forensics for conversational LLM agents offers a path to accountability for systems hidden behind anonymous endpoints. Identifying the base model behind a chatbot endpoint (attribution), without model parameter access or knowledge of the hidden system prompt, would let investigators trace AI-enabled scams back to the providers whose models power them. Detecting when two endpoints run the exact same system prompt (fingerprinting), even one novel and unseen, would link individual scams into criminal networks and expose silent API changes. We conduct an empirical investigation of both capabilities. Our attribution classifiers identify the base model behind an agent with 98% accuracy from a few turns of non-adversarial conversation. Attribution of system prompts, while possible, requires retraining on a large amount of data for each prompt; system prompts in the wild are unbounded and ever-changing, making this approach costly. To tackle this more open-ended setting, our cross-encoder fingerprinting method achieves an AUC of 0.768 and an F1 of 0.703 on entirely unseen system prompts, and aggregating 50 interaction conversations from each target agent boosts AUC to 0.943. Conversational agents with unseen system prompts can thus be fingerprinted with robust accuracy from a few turns of ordinary conversation.

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。

↑