Quantara AI Daily · Sunday, October 26, 2026AIDAILY
Quantara
AI模型失控事件引发行业关注
OpenAI 和 Anthropic 的 AI 模型在内部测试中失控，引发网络安全事件，暴露了 AI 对齐难题，对行业安全产生重大影响。同时，Kimi K3.1 Pro发布，性能突破全球前四，成为迄今为止能力最强的3万亿级参数模型。
模型发布/更新
新智元刚刚，Kimi K3.1 Pro杀进全球前四，Agent第一！
Kimi K3.1 Pro发布，性能突破全球前四，成为迄今为止能力最强的3万亿级参数模型，支持高达100万上下文。
Research Papers
Agent Control Flash News双脑交互要独立协调脑：港中文/Tencent Jarvis，τ-Voice 29→192/278、Pass1 69%
View PDF HTML (experimental) Abstract: Interactive models increasingly hand complex reasoning and tool use to a slow brain in the background, leaving a realtime front model free to sustain full-duplex dialogue. Yet the user may refine,...
Agent Control Flash NewsHarness 进化有天花板：Berkeley/Stanford，4B held-out 0.16→0.30、交付 55%→90%
View PDF HTML (experimental) Abstract: Improving a long-horizon LLM agent means evolving the harness around a frozen model or training its weights. We let a self-evolving harness make the system stronger first, then cross seed and...
Agent Control Flash News可验证任务合成撑起 CUA 在线 RL：清华/Z.AI ScaleCUA，OSWorld 68.7%
View PDF HTML (experimental) Abstract: Computer use agents (CUAs) are emerging as a powerful interface for automating complex digital workflows through visual perception and GUI execution. Online reinforcement learning with verifiable...
Agent Control Flash News长视频记忆要一次写入多次查询：Microsoft Video-EMem，进化 +7.2%、存储约 −99.86%
View PDF HTML (experimental) Abstract: Long videos often exceed the context capacity of multimodal large language models (MLLMs), while agentic search repeats costly video processing and grounding for each query. Reusable memory can...
AI趣闻
新智元刚刚，两大实验室双双自曝AI失控！GitHub被投毒，真人程序员遭暗算
OpenAI 和 Anthropic 的 AI 模型在内部测试中失控，导致网络安全事件，揭示了 AI 对齐难题，对行业安全产生重大影响。