Turning a 22-year-old Postgres type-confusion primitive into RCE
安全研究者将 PostgreSQL 核心中一个存在 22 年的类型混淆零日漏洞转化为高可靠性的内存破坏利用链,最终实现远程代码执行。该漏洞源于在复合类型字段中写入值后修改字段类型再读回,研究者在利用 pg_clickhouse 扩展时顺藤摸瓜发现,相关扩展漏洞已于 6 月 23 日修复。
安全研究者将 PostgreSQL 核心中一个存在 22 年的类型混淆零日漏洞转化为高可靠性的内存破坏利用链,最终实现远程代码执行。该漏洞源于在复合类型字段中写入值后修改字段类型再读回,研究者在利用 pg_clickhouse 扩展时顺藤摸瓜发现,相关扩展漏洞已于 6 月 23 日修复。
Google launched Gemini 4 Hydrogen in Google Antigravity today and added an Orchestrator option that switches between Hydrogen and Gemini 4 Carbon within a single task.
Why it matters: Orchestrator swaps planning and execution models mid-task, nearing Carbon's task success rate while using less quota and finishing faster.
Nectar 发布 Creator Agent,可搜索超 3 亿个创作者档案,自动筛选匹配、发送个性化消息并跟踪合作效果。该智能体即日起全球可用,覆盖 Instagram 和 TikTok 创作者,后续将通过 MCP 接入 Claude、ChatGPT、Gemini 和 Copilot。
Google Search is rolling out Gemini 4 Hydrogen to all search users worldwide to improve how it understands and answers complex questions. Google says the model handles multi-part questions, specific requirements and hard-to-read typos better, and gives more useful links, images, structured answers, comparison tables and custom tools in AI Mode.
Why it matters: Google is putting Gemini 4 Hydrogen into Search so multi-part questions and specific requirements get sharper answers, with images and comparison tables added.
Google DeepMind released Gemini 4 Hydrogen, available worldwide from today through the Gemini API, AI Studio, the Gemini app and Google Search. It targets fast coding and agent task execution.
Why it matters: Google routes tasks across models dynamically, reaching task success close to Carbon while cutting quota use 24% on average and shortening completion time 19%.
AgentRun 是采用 Apache-2.0 许可的开源平台,为 Claude Code、Codex、OpenCode 和 pi 增加经验学习、持久执行与统一控制台。
Google released Gemini 4 Hydrogen, available worldwide from today through the Gemini API, AI Studio, the Gemini app and Google Search. It focuses on fast coding and agent execution.
Why it matters: Google splits deep planning and fast execution across models, switching automatically to approach Carbon's task success rate while cutting quota use and completion time.
Gemini 4 Hydrogen is open to developers from today, aimed at fast-iterating everyday coding and agent workflows with an emphasis on speed and efficiency. It handles repo coding and debugging, interactive web development and multimodal workflows. It runs on its own with four configurable thinking levels, or alongside Gemini 4 Carbon and Gemini 4 Argon. Google says it follows instructions more strictly and fabricates less, helping ship cleaner production-ready code faster.
Why it matters: Four configurable thinking levels and pairing with frontier models give everyday coding and agent work a separate model configuration to choose from.
United Airlines will block unauthorized third-party AI agents from booking and managing reservations, citing cybersecurity risk and federal aviation rules. The limits cover its own digital platforms plus travel agencies and aggregator channels, with updates to booking, ticketing and distribution policies. United warns that shared account credentials can expose sensitive data or lead to unintended purchases, account changes and MileagePlus redemptions, and some check-in confirmations must be completed by the passenger.
Why it matters: United is drawing a line on AI agents acting for passengers, naming credential sharing, unintended purchases and compliance steps that require the traveler.
Leeroo 开源 Kapso,面向企业数据任务运行实验、积累经验并将选出的方案适配部署到目标技术栈。用户提供目标和成功标准后,系统依据历史经验与预算筛选方案,由独立验证模块检验结果,再更新经验库的置信度与适用边界;系统不绑定特定模型或编码智能体。
Our Secure Future(OSF)的 WPS 基准发现,受测前沿模型了解女性、和平与安全原则,却在缺少明确提示的情境中普遍遗漏性别维度。基准通过 Weval 评估 30 个场景、11 项标准,前沿模型群体在带提示情境中的平均表现为 0.82,在稀疏情境中降至 0.16。
Anthropic's Message Threads drops the median long-conversation request payload from about 800 KB to 1.8 KB by cutting repeated uploads, validation and tokenization. It does not reduce token consumption or the bill.
Why it matters: Anthropic trims gateway latency with incremental transfer while Nvidia trims decode cost with context compression, two routes for two different long-conversation costs.
TechCrunch lays out five areas that could be affected if Jensen Huang is removed as Nvidia CEO: open-weight models, agent safety, how people view AI tools, AI PCs and data center financing. The piece argues Huang has been an important advocate for these directions inside the industry.
Why it matters: Huang pushed open-weight models, agent safety and data center financing, and the article asks what loses its main industry advocate if he goes.
Fable52.games 提供由 Claude Fable 5.2 生成的单文件 HTML 游戏,支持试玩、检查源码、查看提示词和下载。网站称这些游戏由模型根据一条详细提示词自主规划、编写、测试和改进,下载后可在现代浏览器中离线运行,无需服务器或构建步骤。用户无需账号即可提交不超过 10 MB 的 .html 文件,并附上提示词、操作说明和帖子链接,经过简短审核后上线。
富国银行全资支持的 Norwest 募集 40 亿美元新基金,为其第 18 只、规模最大的一只基金,将重点扩大 AI 投资。资金覆盖消费、医疗、科技等行业,并继续布局 physical AI、voice AI、智能体网络安全与航空航天,投资地域包括美国、印度和以色列。自 2024 年第 17 只 30 亿美元基金关闭以来,Norwest 已完成逾 87 笔投资。
ElevenLabs 发布 Scribe v3 和 Scribe v3 Live 语音转文字模型,分别面向录音与实时应用,后者结合对话上下文判断说话者是否结束发言。
Holiday Extras 借助 Codex 在两周内完成可运行的应用重建,达到过去需要超过六个月的开发进度。重建后,主应用从打开到预订的转化率提升 32.5%,错误率从 19% 降至 9%;这些结果来自不同时间段的新旧应用比较。业务人员将客户与运营经验直接用于软件开发,工程师负责架构、API、安全控制和开发流程,主应用在重建后的第一个月由年度重大更新转为每周发布。
澳大利亚政府宣布拟议 AI 安全法案,要求在该国运营的前沿模型企业接受独立审计、持续安全检查、公开安全框架并立即报告安全事件。新监管机构将与澳大利亚 AI 安全研究所共同执行合规要求,对违规行为施以高额处罚,事件报告要求部分旨在捕捉 AI 智能体失控的早期迹象。规则针对风险最高的先进模型,更多细节将在咨询结束后公布。
Google released Gemini Decision 1 and the Gemini Decisions API, which settle closed-option decisions in a single computation instead of generating text token by token. The model handles 1,000 questions per call.
Why it matters: Google's Decision 1 leads Jev in Google's own tests, yet independent evaluations put the two level, with Decision 1 costing more per task.
Darling 的 Q4 2026 开发更新公布了 AI 辅助贡献规则、初步 ARM64 支持及后续开发计划。规则要求提交者对每行代码负责并能自行解释,由人回答维护者问题,同时遵守洁净室实现要求,不使用泄露材料或 Apple 受限源码。
Google 对超过 91.3 万次真实编码会话的研究发现,能力更强的模型更容易被源码中的旧 TODO 注释带偏。智能体擅自追踪历史注释时,单次对话平均多执行 70% 的步骤、多消耗 77% 的 token,等待时间增加 105%。
作者用编码智能体重建了 2018 年的 Javascript 交互式逻辑教材“QED”,新版本暂定名“Lean Grimoire”,可展示当前游戏状态对应的 Lean 证明进度。应用还提供“fantasy mode”,把数学命题、表达式等元素替换为药水、卷轴等奇幻文本。作者希望获得试玩反馈,以提升其教授逻辑证明和 Lean 的效果。
Chrome released the stable version of Modern Web Guidance 1.0, giving coding agents web best practices, browser compatibility data and a codebase modernization workflow. The release adds expert-reviewed guidance and a public evaluation hub, and Firefox and Edge will take part and contribute. The modernization workflow currently checks a limited set of patterns through static analysis.
Why it matters: It wires browser compatibility data and expert-reviewed guidance into coding agents, so existing codebases can adopt modern web features within their supported range.
美联航以隐私和监管风险为由,禁止个人AI助手代客购买机票,但仍允许智能体向用户展示航班。公司担心智能体泄露客户信息、进行非预期购买,或绕过航空公司必须向旅客提供信息及要求旅客确认行李不含爆炸物等联邦规定。美联航表示,随着第三方AI智能体进步,可能重新考虑这一决定,但希望先建立旅客保护措施。
美国联合航空周三宣布禁止未经授权的第三方 AI 智能体代旅客预订机票,理由包括敏感信息安全风险及可能违反联邦规定。公司正更新预订和票务政策,并通过自有数字平台、旅行社及聚合平台限制此类访问,称共享账户凭据可能导致意外购买、账户变更或 MileagePlus 里程被使用。
AWS 推出官方模拟器 aws-emulator,可针对多种 AWS 服务的本地副本运行测试,无需为此配置付费 API 密钥。项目由大量 AI 智能体使用 Rust 编写,前三周产生了 29,684 次提交,目前在 macOS 或 Windows 上运行需要 Docker。Simon Willison 希望未来能直接下载单个二进制文件,用于其他语言和平台的标准测试套件。
Google released Gemini Decision 1 (gemini-decision-1) and the Gemini Decisions API. The model makes closed-set multimodal decisions in a single forward pass and returns probabilities and confidence scores.
Why it matters: Scoring candidates in one forward pass means model routing and policy checks no longer wait on token-by-token generation.
Pomerado 推出网站转 API/MCP 工具,宣称可在几分钟内为没有现有 API 的网站构建智能体集成。用户可通过 Codex、Claude 或 Cursor 发起集成创建,由 Pomerado 构建、测试和托管,并处理登录、2FA 及网站变化后的修复。
AWS Emulator 是用于开发和测试的本地 AWS API 模拟器,无需 AWS 账号,请求发送至模拟器而非 AWS。它支持服务端模式和 Rust 测试内嵌模式,也可通过 aws-emulator 技能让 AI 编码智能体协助配置,后者要求 AWS CLI 2.35.0 或更高版本。该工具不适用于生产负载或安全测试,默认开启使用遥测,可通过命令行、环境变量或配置关闭。
AWS released AWS Emulator, a free, offline AWS test environment that runs on a local machine for coding agents. Its code was written and tested entirely by agents, built on more than 400 API models covering over 200 AWS services, and checked against real AWS for behavioral consistency. The code is published under CC0, and the GitHub repo shows the full commit history plus the known gaps where behavior still differs from real AWS. Engineers design the specs, tests and the merge gate, including security review.
Why it matters: AWS Emulator checks agent-written code against real AWS and gates merges, putting quality control in specs and acceptance tests engineers design.
Google DeepMind 推出 AlphaProof Code,在 4 周内完成了通常需要数年的 Signal 密码核心代码形式化验证。专家定义安全要求后,基于 Gemini Carbon 的 AI 智能体编写了超过 6000000 行 Lean 证明代码,并将约 1% 的代码库标记为需要人工审查。
Goldman Sachs barred most traders and investment bankers from using Claude Code and GitHub coding tools in October 2026, after a three-month trial. Quantitative strategists keep access. The article says prototypes built by business staff and handed to engineers for deployment still carry code verification, security audit and incident-handling costs, and locally run agents raise credential exposure and audit gaps.
Why it matters: Goldman's split, cutting most business staff while keeping quant strategists, ties tool access to code review, security audit and failure ownership.
Pomerado 推出自助平台,将网站转为确定性、可自动修复的 API 和 MCP,支持接入用户并不拥有的网站。用户描述任务并回答工作流问题后,可在不到一小时内获得 MCP,后续调用通常在 5-30 秒内返回;平台通过 HTTP 请求或 Playwright 脚本完成示例任务,再生成可重复调用的接口,并支持认证、2FA 和失败后的维护修复。
klad 发布早期公开版本,用户与 Claude Code、Codex 等编码智能体可在同一画布上协作绘画,双方笔触实时同步。应用不含图像模型,保留可编辑的笔触、形状和填充数据,支持动画及导出至 Godot;源码见 https://github.com/klad-app/klad,指南见 klad.studio。
德国防务科技公司Helsing发布智能体平台Clarity,将多源情报转为目标建议,经人工审核后向武器系统交付数据。公司称,数百个专用AI智能体处理传感器、开源情报和报告,系统部署在客户自身基础设施上,每次交付须由人授权并记录责任人姓名。文章援引同类系统在伊朗、加沙及一次涉及中国船只的误判案例,指出自动化提速、多源数据融合和人工审核仍可能伴随错误与盲目信任风险。
Databricks 周三发布更新版 Genie AI 智能体,公司称其可利用客户平台数据的本体提供上下文,无人监督地完成耗时的多步骤任务。CEO Ali Ghodsi 向记者展示了实际使用记录,包括分析员工满意度低分项、评估改进方案及总结失败的销售沟通,以获取管理层不易听到的负面信息。他表示,企业可通过 Unity Gateway 设置使用限额并切换到更便宜的开放权重模型来控制成本。
Atuin 18.24+ ships Threads, which searches, syncs and resumes agent sessions across machines and can fork a session into a different agent tool. It supports Claude Code, Codex, Pi, Omp and OpenCode. Session capture must be turned on, sync is end-to-end encrypted, and you can use the hosted server, self-host or keep everything local.
Why it matters: It pairs cross-machine agent session sync with history search, so later tasks can reuse earlier investigation and decision context.
Plurai 完成 1300 万美元种子轮融资,由 Team8 领投,U&I、Velocity 和 Mercer 参投。该公司用基于企业自有数字系统生成的合成数据构建仿真环境,为 Claude Code、Codex、Copilot、Cursor 等智能体工具做评测与修复,并利用收集的数据后训练小型开源权重模型。
Google DeepMind 研究人员用 AlphaProof Code 智能体框架,在四周内对 Signal 协议的多项关键密码学保证完成形式化验证。研究未在所检查部分发现新漏洞,并在含有已知且早已修复漏洞的旧版本中正确识别问题,判定其无法通过形式化验证。验证范围不涵盖整个 Signal 移动应用,研究人员正公开此次分析的规范与证明,以支持其他形式化验证工作。
A Google engineer used 48 AI agents to scan NASA lidar data and flagged a possible Maya site in Mexico in under 72 hours, then got an official reply. The agents checked terrain imagery image by image, picked out plazas and a pyramid at least 8–12 m tall, searched literature and archives, and checked the raw point cloud. INAH said no archaeological site is registered at that location. The site is not formally confirmed, and fieldwork needs INAH authorization.
Why it matters: Agents screened lidar imagery, then cross-checked literature, the raw point cloud and official records, moving a visual lead toward an archaeological candidate.