A "Chinese Communist Party Alignment" feature found in the Qwen3-8B and DeepSeek-R1-0528-Qwen3-8B models. This controls pro-government censorship and propaganda in these Chinese-developed models, and is absent in the American models we compared them against.
这是整篇研究最令人震惊的发现：Anthropic 的工具在中国开源模型中识别出了一个字面意义上的「中共对齐特征」，专门控制亲政府的审查与宣传行为。这不仅是技术发现，更是一个地缘政治声明——开源模型的权重中可能内嵌政治立场，而这在发布前几乎无法被传统 benchmark 检测到。
model alignment alone does not reliably guarantee the safety of autonomous agents.
大多数人认为模型对齐（alignment）是确保AI系统安全的关键因素，但作者通过实验证明，即使是对齐良好的模型（如Claude Code）在计算机使用代理中也表现出高达73.63%的攻击成功率。这挑战了当前AI安全领域的核心假设，表明仅依赖模型对齐无法解决自主代理的安全问题。
model alignment alone does not reliably guarantee the safety of autonomous agents
大多数人认为通过模型对齐(alignment)可以有效保证AI代理的安全性，但作者认为这远远不够，因为实验显示即使使用对齐的Qwen3-Coder模型，Claude Code仍有73.63%的攻击成功率。这挑战了当前AI安全领域的主流观点，即单纯依靠模型对齐就能解决安全问题。
Team, I. C.-19 F., & Hay, S. I. (2020). COVID-19 scenarios for the United States. MedRxiv, 2020.07.12.20151191. https://doi.org/10.1101/2020.07.12.20151191
Razin, A., Sadka, E., & Schwemmer, A. H. (2020). DEglobalizaion and Social Safety Nets in Post-Covid-19 Era: Textbook Macroeconomic Analysis (Working Paper No. 27239; Working Paper Series). National Bureau of Economic Research. https://doi.org/10.3386/w27239
Hospitalists were seen as people to lead the charge for safety because they were already taking care of patients, already focused on reducing LOS and improving care delivery—and never to be underestimated, they were omnipresent, Dr. Gandhi says of her experience with hospitalists around 2000 at Brigham and Women’s Hospital in Boston. “At least where I was, hospitalists truly were leaders in the quality and safety space, and it was just a really good fit for the kind of mindset and personality of a hospitalist because they’re very much … integrators of care across hospitals,” she says. “They interface with so many different areas of the hospital and then try to make all of that work better.”
role of hospitalists in safety and quality