[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-82af5716-322e-46cf-9de5-b85e8cdd5712":3},{"id":4,"title":5,"summary":6,"content":7,"original_url":8,"source_id":9,"tags":10,"translations":20,"published_at":21,"created_at":22,"modified_at":23,"is_published":24,"publish_type":25,"image_url":7,"view_count":26},"82af5716-322e-46cf-9de5-b85e8cdd5712","微软首推安全专用模型 MAI-Cyber-1-Flash:小模型+多智能体,把漏洞挖掘成本砍半","7月27日,微软 AI 正式发布 MAI-Cyber-1-Flash,这是微软首个面向网络安全场景专门训练的语言模型,脱胎于 MAI-Thinking-1 家族,主打在大型代码库里识别高难度漏洞。模型并非以单点性能取胜,而是被嵌入到自家的多智能体漏洞识别与修复框架 MDASH 中:90% 的常规任务交给 MAI-Cyber-1-Flash 处理,剩下 10% 真正棘手的样本再交棒给 GPT-5.4。结果是,这套 MDASH + MAI-Cyber-1-Flash 的组合在 CyberGym 基准上拿到 95.95%,比 Mythos、Claude 等对手高出约 12 个百分点,同时相对纯 GPT-5.4 的多模型堆栈降低 50% 成本。\n\n更值得注意的是训练数据的护城河。微软每天承接超过 100 万亿条来自身份、终端、云、网络的安全信号,横跨 MSRC 漏洞库与 160 万企业客户的真实攻防数据,这种规模的历史经验几乎无法被复现。模型训练上,微软专门用安全优先的校准,红队 + 第三方独立评估,再叠加企业级 RBAC、租户隔离、加密、沙箱无外网等护栏。\n\n同一天,微软还推出了 agentic 安全系统 Perception,让多个 AI 智能体在 SOC 里持续监控、打补丁、闭环威胁,未来也将接入 MAI-Cyber-1-Flash。这标志着大厂打法正在切换:不再追求单一万能超大模型,而是训练小而专的小模型 + 多智能体编排 + 数据闭环,在垂直领域用极致性价比替代通用前沿模型。对安全行业来说,这是一次范式升级——攻防两端同时被 AI 加速,Defender 端的成本如果率先被打下来,主动防御就有可能从奢侈品变成标配。",null,"https:\u002F\u002Fmicrosoft.ai\u002Fnews\u002Fintroducing-mai-cyber-1-flash-inside-mdash\u002F","9f1ed564-b7ac-4084-b1a8-4d129772db34",[11,14,17],{"id":12,"name":13,"slug":13,"description":7,"color":7},"e676a5cf-1f24-472f-a765-86fa21a1bc3c","ai-model",{"id":15,"name":16,"slug":16,"description":7,"color":7},"5e628969-6d2a-437f-998a-104e4b16cfb1","ai-progress",{"id":18,"name":19,"slug":19,"description":7,"color":7},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",[],"2026-07-28T01:00:00Z","2026-07-27T20:03:45.078519Z","2026-07-27T20:03:45.078528Z",true,"agent",26]