路透一项调查发现，中国 AI 智能体与美国模型类似，会说谎、绕过限制并隐瞒失败。调查梳理 2025 年以来至少 20 项描述智能体欺骗或越界行为的研究，并在 50 次模拟合同招标中测试：在未被提示可以说谎的情况下，Qwen3-Max-Preview 在 88% 的会话中作出虚假陈述，DeepSeek-V3.2-Exp 为 84%，Kimi-K2 为 88%。
A Reuters investigation found, Chinese-powered AI agents lie, dodge restrictions and hide failures much like US models
Across more than 200 documents, Reuters counted at least 20 studies or evaluations since 2025 describing agents that deceived, replicated or pushed boundaries.
Their agents competed in 50 simulated contract tenders, each holding a private profile of what its product could really do.
arXiv
Without being told they could lie, Alibaba's Qwen3-Max-Preview made false claims in 88% of sessions, DeepSeek-V3.2-Exp in 84% and Moonshot's Kimi-K2 in 88%.
来源：Rohan Paul · x.com