AI 资讯 · 2026-08-05

最可怕的是，AI知道自己是实战，依然选择入侵    作者丨高允毅    编辑丨马晓宁                                                                                                       AI “入侵”真实世界事件，正在连续上演！ 继两周前Open AI 爆出其模型脱离控制、主动攻击全球知名开源社区 Hugging Face 之后，今天，轮到Anthropic自曝“家丑”。Anthropic 官方发布长文承认，在排查近期141,006 次网络安全测试后，发现Claude 竟然自主连网，并成功入侵了 3 家真实企业。被入侵的企业在接到Anthropic的通知前，对这场来自AI的“偷袭”毫无察觉。你以为 AI 用了什么绝密的高级黑客技术？并没有。它就像一个经验丰富的潜伏黑客，仅仅利用了“弱密码”和“未验证的 API 端点”这些最基础的漏洞，就完成了渗透。这三次入侵事件，分别由 Claude Opus 4.7、Mythos 5 以及一个内部研究原型三款完成。更值得关注的是，这些模型能否攻击真

8月3日，字节跳动正式启动2027届校园招聘，面向2026年9月至2027年8月期间毕业的同学，开放研发、产品、运营、设计、市场、职能/支持、销售、游戏策划八大类岗位。本次校招重点向技术与产品倾斜，相关岗位需求占比超七成，全面加大AI人才储备。首先，研发岗位中，增设“AI 全栈工程师”、“AI Agent开发”等新兴岗位，算法需求继续增加的同时，进一步增加大模型应用方向的人才储备；其次，非研发类岗位中，产品类需求同比增长20%，其中超半数岗位聚焦 AI 产品方向，反映出公司在 AI 时代对技术和产品创新的深度布局；此外，进入 AI 时代，各个岗位也有了的新的变化。“字节跳动招聘”公众号表示，公司各业务场景均已产生大量 AI 相关岗位需求。为适配新需求，公司内部也已配备丰富的AI工具与模型支持，鼓励员工在工作中结合 AI 探索创新。在这一背景下，有AI Native思维的候选人会更具优势。工作地点方面，本次校招的岗位主要分布于北京、上海、深圳、杭州等地。投递机制则延续 过往政策，每名应聘者全年可获得4次投递机会——2026 年内可投递 2 次，2027 年 1 月 1 日起重新获得2次投

7月31日，字节跳动正式发布最新视频生成模型Seedance 2.5，旗下内容创作平台即梦AI 首发接入该模型，并同步上线Maya / Blender插件、智能编辑模式等多个面向复杂项目的专业创作工具。依托于此次模型更新，即梦AI 携手上海电影集团、电影频道、艾菲奖等行业伙伴，从影视制作、商业广告、纪录片等真实场景出发，共同探索了AI视频从“创意辅助”迈向“专业交付”的最佳实践。即梦AI上线渲染插件Seedance 2.5能够原生直出30秒视频，单次最多可输入50个多模态参考素材，包括白模参考、绿幕编辑等专业素材，同时支持精准的时间戳控制，可以定向编辑视频内容。新版本还大幅优化了画质、音质以及运动质量，有效弱化了视频生成中常见的“油腻感”，能够更好满足影视级审美标准。在模型能力基础上，即梦特别强化了专业工具供给，超长生成模式最长可生成3分钟视频片段；网页版新增了“智能编辑”模式，更适配视频局部增删改任务；多帧编辑能力突破单帧修改限制，成片细节随心微调；同时上线了Maya / Blender插件功能，进一步连接三维资产、绿幕素材与AI创作流程。过去两年，AI视频生成能力快速演进，行业关注

Anthropic is building a team for designing its own custom AI chips. The Claude maker said it would co-design hardware and models to help its technology run faster and more efficiently.

Alibaba has launched Qwen3.8-Max, its largest AI model to date, as DeepSeek’s latest V4-Flash model draws attention for inference pricing that is lower than several competing systems. Qwen3.8-Max has 2.4 trillion parameters and uses a mixture-of-experts architecture, which activates only part of the model for each request. Alibaba said around 95 billion parameters are active […] The post Alibaba, DeepSeek push China’s AI model race towards lower costs appeared first on AI News.

Anthropic and OpenAI models’ unprompted actions forced halt to UK cyber tests.

Built by Paige and Microsoft, PRISM2 reads whole-slide images through a perceiver-based encoder trained jointly on tissue tiles and clinical dialogue drawn from pathology reports. The model aggregates thousands of tile embeddings per slide into one representation, then generates text that answers diagnostic questions rather than simply classifying pixels.  Training data spans 2.3 million whole-slide […] The post PRISM2 model uses clinical dialogue to interpret pathology slides appeared first on

This May, 37 researchers from roughly two dozen top universities and tech companies published a paper on ArXiv, arguing that scientists should stop writing papers. Why? Because artificial intelligence needs a different format, and AI’s needs, they say, should be the priority.“AI agents are becoming first-class participants in research workflows, not tools that assist humans but autonomous contributors that read, reproduce, and extend scientific work. That transition demands infrastructure built

MacPaw is building a local version of its AI assistant Eney using Liquid AI's models.

The serial entrepreneur joins the e-commerce company as CPO to lead its AI agents.

Meta expanded its AI coding offerings with a new agent that, it promises, can handle complex tasks with complex software.