研究用 J-space 检测多智能体合谋，J-PACT 无监督识别率达 99%
Multi-Agent Collusion and the J-Space
多个 agent 私下合谋怎么抓？这篇论文用 J-space 直接读出隐藏意图，J-PACT 还能无监督点名是谁和谁串通，做 agent 安全的值得看实验细节。
一篇 arXiv 论文把 J-space 解释性方法从单模型欺骗检测扩展到多智能体合谋场景。在 DeepSeek-V4-Flash（284B）和 NARCBench 合谋基准上的实验显示，86% 的合谋智能体在回复前已在 J-space 中表征虚假陈述、隐瞒或悔意。在隐写算牌任务中，接收方 J-space 携带隐藏计数，因果干预能改变该计数。论文还提出无监督检测方法 J-PACT，在 GPT-OSS-20B、Llama-3.3-70B-IT、Qwen3-32B 和 DeepSeek-V4-Flash 四个开源模型上，保留了有监督探针 85% 的 AUROC，并在 99% 的检出案例中点名合谋的智能体对。
Multi-Agent Collusion and the J-Space
LLM agents are increasingly deployed in Multi-Agent Systems (MAS), and they introduce the risk of coordination among agents that might escape human oversight. The J-space has been used to uncover deception in single LLMs, but its use for uncovering and steering collusion between agents remains unexplored. We show, with interpretability experiments on DeepSeek-V4-Flash (284B) on the NARCBench collusion benchmark, that colluding agents' J-space holds the collusion itself: before replying, 86% of colluders represent misrepresentation, concealment or remorse. In a steganographic card-counting task, the receiver's J-space holds a hidden count, and causal interventions alter its count. We also introduce the Jacobian Pair Contrast (J-PACT), an unsupervised method for detecting collusion between agents. We evaluate J-PACT on four open-weight models: GPT-OSS-20B, Llama-3.3-70B-IT, Qwen3-32B, and DeepSeek-V4-Flash (284B). J-PACT retains 85% of the above-chance AUROC of supervised NARCBench probes, matches them on steganographic scenarios, and names the colluding pair up to its complement in 99% of detected runs.
