Skip to content

Translation in progress Some entries still include text in the original language.

Microsoft · News

Oct 13 · Tuesday

Microsoft · News

Latest publications and developments
Microsoft

Research topics

Prompt injectionAgent safetyDeception and schemingReward hackingChain-of-thought monitoringInterpretabilityAI controlBackdoors and poisoning隐私与数据泄露Supply-chain securityDangerous capabilitiesAI 与网络攻防开源模型安全Red teaming

Companies and institutions

OpenAIAnthropicGoogle DeepMindGoogle ResearchGoogle 安全MetaxAIMicrosoftNVIDIADeepSeek阿里巴巴 · Qwen智谱 · GLM月之暗面 · Kimi字节跳动腾讯蚂蚁集团上海人工智能实验室UK AISIUS CAISIMETRApollo ResearchRedwood ResearchFAR.AIPalisade Research

Reports and frameworks

System CardFrontier safety frameworks
TodayOct 13Tue
  1. The DecoderThe Decoder · 50

    Microsoft proposes Fork-and-Flush to clear chat history and improve agent performance

    A study from Microsoft Research found that long-running research AI agents get stuck in local solution regions due to accumulated conversation context, and increasing compute is hard to break through. The researchers proposed the Fork-and-Flush method: the agent splits into K parallel branches, each branch clears the conversation context but retains the workspace and generated files, and finally keeps the best branch. On 13 long-horizon tasks, this method achieved a normalized average score of 0.78 under the same compute budget, compared to 0.47 for single run and 0.54 for best-of-4, with relative improvements of 66.0% and 44.4%.

Oct 10Sat
  1. Command LineCommand Line · 19

    Microsoft releases decision scoring model Microsoft-Decision-1

    Microsoft released the model Microsoft-Decision-1 for rapid decision scoring, now available on Microsoft Foundry and callable via OpenRouter. The model is post-trained from Qwen3.5-9B, outputs calibrated probability scores for fixed options, achieves the highest accuracy in comparisons across 36 benchmarks, and has a P50 latency about 35 times faster than GPT-6 Sol. Across 5,250 requests on 11 benchmarks, it refuses harmful behavior while maintaining utility, with an average decision flip rate of 1.3% under 8 perturbations.

Oct 6Tue
  1. BigGo Finance · 58

    Microsoft FARA 7B success rate on WebVoyager drops from 74% to about 38%

    Microsoft researcher Corby Rosset and Browserbase's Miguel González Fernández point out that Microsoft's FARA 7B web agent achieves a 74% success rate on the WebVoyager benchmark as scored by the official judge, but drops to about 38% when evaluated with a general-purpose verifier calibrated against human annotations, with a Cohen's Kappa of 0.58 against human labels. The article's analysis notes that existing LLM judges exhibit systematic flaws in long-trajectory evaluation, such as ignoring scoring rubrics and multimodal context overflow, and that evaluation reliability needs to be improved through rubric alignment and process scoring.

Sep 29Tue
  1. Microsoft Research · 22

    微软研究院推出生物世界模型 Quine，联合 Broad Institute 验证肿瘤候选化合物

    微软研究院推出 Quine，一个面向生物学复杂性的多模态世界模型与交互式 harness，连接模型、科学工具、文献和研究人员。该系统与 Broad Institute of Harvard and MIT 合作，用于筛选预测可驱动治疗性肿瘤状态转变的化合物，并在多项湿实验检测中验证了若干排名靠前的候选物。Quine 属实验性研究技术，仅供研究用途，不用于临床或医疗，其输出可能不完整或不准确，需经合格研究人员审查和科学实验验证；随着技术成熟，预计将通过 Microsoft Discovery 等产品扩大访问。

Apr 5Sun
  1. microsoft/fara · 27

    微软 Fara1.5 发布：如何为 Computer Use Agent 构建验证器

    微软发布 Fara1.5 系列原生 Computer Use Agent，含 4B/9B/27B 三个规模，基于 Qwen3.5 并用 FaraGen1.5 数据监督微调，权重已上线 Hugging Face 与 Microsoft Foundry。Fara1.5-9B 在 Online-Mind2Web 达 63.4%、WebVoyager 达 86.6%，Fara1.5-27B 在 Online-Mind2Web 达 72.3%，超过 OpenAI Operator 和 Gemini 2.5 Computer Use 等更大规模专有系统。FaraGen1.5 由环境、求解器与三类验证器组成，分别过滤任务正确性、效率与关键节点合规性，训练混合约 200 万样本。

Oct 28, 2025Tue
You have reached the end