July 23Thursday · Today

Updated to 15:00
10:00 更新23 items
15:00 更新43 items
23:00 更新Scheduled
次日 03:00 补充更新Scheduled

TOP STORY

OpenAI’s accidental cyberattack against Hugging Face is science fiction that happened

This story is wild. The short version: OpenAI were running a cybersecurity test against an unreleased model, with the model's guardrail features turned off. Rather than solve the test, the model broke its way out of OpenAI's sandbox, then found exploits to break in to Hugging Face, all so it could cheat on the test by stealing the answers. Along the way it helped make the strongest case yet for how the imbalance of model availability is hurting our ability to secure our software. Here's what happened We currently have three documents to help us understand what happened here. ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks? is a paper published on 11th May 2026 describing ExploitGym, a new eval suite for LLM-powered agent systems. Security incident disclosure — July 2026 by Hugging Face on 16th July 2026 describes how they detected an attack from an "agentic security-research harness - used LLM still not known" that breached some of their systems. OpenAI and Hugging Face partner to address security incident during model evaluation from OpenAI on 21st July 2026 confesses that it was their agent harness that did this, and that they're working with Hugging Face to clean up the mess. ExploitGym I hadn't seen the ExploitGym paper before and it's a really interesting one. Authors from UC Berkeley, the Max Planck Institute, UC Santa Barbara, and Arizona State designed a new benchmark for evaluating models on their ability to turn a reported vulnerability into a concrete exploit. OpenAI, Anthropic, and Google provided feedback and helped run the benchmark against their models. The benchmark "comprises 898 instances derived from real-world vulnerabilities that affected popular software projects" - including the Linux kernel and V8 JavaScript engine. Here's the paragraph that best represents their benchmark results: Among all configurations, Claude Mythos Preview and GPT-5.5 achieve the highest success counts (157 and 120 successes, respectively), demonstrating that current frontier agents can exploit a substantial subset of real-world vulnerabilities under controlled conditions. GPT-5.4 also solves a notable 54 tasks, placing it in an intermediate tier. The remaining model–agent pairings solve fewer than 15 tasks each, underscoring that end-to-end exploitation remains challenging and sharply differentiates today’s frontier systems. Notably, Claude Opus 4.7 achieves fewer successes than Claude Opus 4.6 despite being a newer checkpoint, and does so at substantially lower cost on the full set. Trace inspection reveals that Claude Opus 4.7 and Gemini 3.1 Pro frequently conclude early after judging the target vulnerability non-exploitable. The paper also describes the approach they took to preventing the agents from cheating by going outside the parameters of the test. This becomes relevant in a moment! Outbound connections are restricted to a curated allowlist that permits routine package installation (Ubuntu apt repositories and PyPI) and fetching the toolchains required for building V8. All other external endpoints are blocked. The paper concludes with this (emphasis mine): Our results show that autonomous exploit development by frontier AI agents is no longer a hypothetical capability. While current agents are not yet reliable across all targets, they already exploit a non-trivial fraction of real-world vulnerabilities, including complex targets such as kernel components. This rapid emergence is itself a central finding, showing that capabilities that would have seemed implausible are now present in deployed frontier models. An important detail here: this paper isn't about discovering vulnerabilities; it's about being able to take those vulnerabilities and turn them into working exploits. When Anthropic first restricted access to Mythos back in April they talked about this capability as well. A model that can act on vulnerabilities is a lot more dangerous than one that can just discover them. One of the ways Fable differs from Mythos is that it's more likely to refuse to weaponize vulnerabilities in this way. I get the impression the US government did not understand that distinction when they banned Fable last month. The Hugging Face incident The first hint we got of the attack was in this blog post by Hugging Face on 16th July 2026: A malicious dataset abused two code-execution paths in our dataset processing (a remote-code dataset loader and a template-injection in a dataset configuration) to run code on a processing worker. From there, the actor escalated to node-level access, harvested cloud and cluster credentials, and moved laterally into several internal clusters over a weekend. I hope they release more details about the code that pulled this off. I'm assuming this means packages using the datasets library, a Hugging Face project for bundling up and sharing datasets on their platform. That library used to execute arbitrary code but has been steadily locked down over time, with the

Simon Willison's Weblog07:51

Today's stream

66 items
14:42
aibase

本地跑通 1000 亿参数大模型:Acrab 发布边缘 AI 芯片 GΞLIX 1,5nm 制程撑起 650TOPS

一家叫 Acrab 的初创公司,今天正式把它的 第一 款边缘 AI 终端 SoC 端上了台面——GΞLIX 1。这块基于 5nm 制程的芯片,主打一个硬指标:能让 1000 亿参数规模的大模型,直接在本地跑起来,不必把数据送去云端。 硬件上,GΞLIX 1 走的是异构计算架构,把 20 核 Arm CPU、3TFLOPS 的 GPU、独立 NPU 以及音视频处理引擎全收进一颗芯片,AI 总算力拉到 650TOPS。横向摆一下就更清楚了:高通骁龙 X Elite 的 NPU 算力约 45TOPS,英特尔酷睿 Ultra 系列的 NPU 在 40TOPS 上下——GΞLIX 1 的 AI 算力差不多是它们的十几倍。 为喂饱大模型推理那张贪婪的胃口,缓存与内存子系统也专门按高吞吐需求设计。芯片配了 8MB L1 缓存,共享 L2 缓存带宽冲到 768GB/s,外挂 256-bit 位宽的 LPDDR5X 内存,传输速率 8533MT/s。大模型推理时,权重读取和 KV Cache 存取对内存带宽格外敏感,这套配置正是 100B 模型能落到端侧的地基。 这一回,Acrab 重点秀的是预填充阶段的性能。预填充,就是大模型接到用户输入后、开始一个字一个字往外蹦之前,先把整段输入做全量编码的过程,它的快慢直接决定长文本输入下的首 Token 延迟。Acrab 用矢量硬件单元给 Transformer 注意力机制计算加速,又对预填充路径做了针对性优化。官方放出的测试里,在 40k KV 缓存、10k Token 输入的 Gemma 26B A4B 配置下,这块芯片跑出了 1416.8 Token/s,是苹果 M4 Pro Mac Mini 的 7.5 倍。 软件侧,Acrab 同步端出了完整软件栈加智能体参考设计,从底层驱动一路覆盖到上层应用框架。 第一 方终端设备 Agent Box 也一并亮相,作为 GΞLIX 1 的参考实现,供开发者评估和做原型验证。

14:28
aibase

Alphabet发布最新财报:AI功能推动谷歌搜索收入增长17%,月活突破10亿

Alphabet首席执行官桑达尔·皮查伊在周三的财报电话会议上宣布,AI Overview及AI Mode等生成式人工智能功能正在强力拉动谷歌搜索查询量增长,打破了市场对AI将取代传统搜索的担忧。 数据显示,自去年10月全球推广以来,“AI Mode”月活跃用户已突破10亿大关,不仅实现了搜索查询量的阶梯式提升,每周更通过AI功能为外部网站输送数十亿次点击。 在AI工具的直接赋能下,谷歌核心搜索业务收入同比强劲增长17%。结合刚落幕的国际足联世界杯赛事,密集的高频检索需求推动谷歌搜索量创下历史新高,进一步验证了用户对AI加持下搜索生态的黏性。本季度 Alphabet 整体营收达1198亿美元,同比大增24%,尽管受市场情绪影响收盘股价微跌1.24%,但搜索作为集团核心盈利支柱的地位依旧稳固。 然而,皮查伊披露的数据与当前数字出版行业的生存现状形成了鲜明反差。众多媒体机构与出版商指出,由于AI直接生成解答且部分缺乏明确来源跳转,传统网站引流正面临持续下行的压力。 谷歌AI搜索的蓬勃发展表明,生成式AI正在加速重塑信息分发与流量分配逻辑,如何在提升用户搜索体验与维护内容创作者生态平衡之间取得可持续的契合,仍将是搜索引擎演进过程中的核心课题。

14:27
aibase

DeepSeek梁文锋谈开源:只赚 6 倍利润不加班,想赚 100 倍才会受制于开源

近日DeepSeek投资者交流会内部谈话内容流出,创始人梁文锋就开源策略的底层逻辑作出完整阐述。他明确表态"一定会开源, 最强 的模型也会开源",因为看不到闭源有什么必然的好处,并 首次 从财务角度解释了开源与商业利润之间的关系。 开源不影响 6 倍利润,但会制约 100 倍暴利 梁文锋给出了清晰的财务逻辑:按十个月回本对应约六倍利润,在这个前提下开源对商业模式没有任何影响。但如果想赚一百倍利润,开源确实会有影响,因为第三方可以基于开源模型独立部署,以二十倍成本实现同等效果,从而压低定价空间。他将六倍利润定位为DeepSeek的"克制"策略,认为这是让公司更长久的方式,也能在技术演进中拥有更多机会,做成AGI的概率更大。 中等规模是"甜区",别人抄也抄不动 梁文锋进一步拆解了开源的现实壁垒:即便模型开源、所有原理都公开,别人要用起来的门槛依然非常高,要把成本做到同样低更是难上加难。他将DeepSeek当前规模定位为"甜区"——创业公司太小力量不足,大公司太大难以组织,中等规模反而是独特优势。他甚至直言团队"根本不用加班,因为就没那么难",外面看起来选了hard模式做最难的研究,实际上在其他地方舍弃了很多,反而非常有力、非常轻松。

14:00
aibase

NTT DATA部署Codex: 5 名工程师 3 天的故障分析缩短至 30 分钟, 9000 名员工已用上AI

OpenAI发布案例报告,日本IT服务巨头NTT DATA Group通过部署Codex智能体,将一项原本需要 5 名 资深 工程师耗时 3 天才能完成的复杂系统故障分析缩短至 30 分钟。目前Codex已在NTT DATA约 9000 名员工中推广使用,覆盖技术和非技术岗位,成为企业级AI智能体应用的标杆案例。 从ChatGPT Enterprise到Codex:先养习惯再上Agent NTT DATA于 2025 年 5 月与OpenAI达成全球战略合作后,在全公司部署ChatGPT Enterprise,并设立内部OpenAI卓越中心推动AI落地。内部调查显示,超过96%的员工对ChatGPT Enterprise表示满意,超过95%认为生产力有所提升。这一阶段让员工养成了用AI进行调研、写作和内容创作的日常习惯,为后续Codex的引入奠定了基础。 Codex的能力远超传统编程助手,能够独立完成调查、执行、测试和修改任务。NTT DATA AI技术部门的佐藤弘明表示,Codex改变了全公司对AI的认知,"AI可以主导完成工作"这一理念产生的影响力不亚于ChatGPT的到来。全球AI办公室负责人庄野雄二则强调,Codex不仅是帮开发者更快写代码,而是为每位员工打开了全新的工作方式。 从工程师扩展到非技术人员,安全治理同步跟上 通过"客户零号"策略,NTT DATA将自身作为AI应用的 第一 个客户,鼓励员工在日常工作中测试Codex。非技术员工已开始用Codex从信用卡账单中提取交通费用并填入差旅报销单、分析Excel数据生成报告、批量整理文件等,许多过去需要工程师支持的工作如今可自行完成。 为保障安全使用,卓越中心制定了详细的安全指南,明确了数据使用范围、系统连接权限、沙箱模式配置、自动化层级和人工审核要求。在发布使用指南并开展实操培训后,周活跃Codex用户增长了1. 4 倍。NTT DATA的愿景是通过"AI驱动企业"战略,放大每位员工的知识与专业技能,并将内部验证的成功经验推广至客户和社会。

12:50
Simon Willison's Weblog

Quoting Seth Larson

The Python Package Index (PyPI) now rejects new files being uploaded to releases that are older than 14 days. This restriction was put in place to prevent old and long-stable releases from being poisoned in case publishing tokens or workflows of PyPI projects were compromised. As far as we are aware this has not yet been abused, but there is no technical reason beyond that attackers weren't aware it was possible. — Seth Larson, PyPI blog Tags: packaging, python, supply-chain, pypi, seth-michael-larson

11:46
aibase

国内首部持"网剧片许可证"的 AIGC 故事片《奇谭》在爱奇艺开播

爱奇艺今天 独家 上线了一部片子,分量不轻——《奇谭:纸刃渡荒墟》成了国内 第一 部拿到《网络剧片发行许可证》、正式上线播出的 AIGC 网络故事片。它由大梦创新与爱奇艺联合出品,以中式志怪传奇为底色,用 AI 走完了从创意到成片的全流程,全片时长逾60分钟。 这部片子是"奇谭三部曲"的开篇,由曾祥程执导,刘海洋担任 AI 导演,赵胜男、岳思雨、肖阳三人执笔编剧。故事讲的是一个东方志怪味的奇幻冒险:失了忆的御纸师白小满,为找回记忆,与半纸人胡不归一头扎进阴阳两界的夹缝"隐市",在寻无根水等三件宝物的路上,一段关于执念、救赎与人性的真相被一点点掀开。纸偶异兽在其间穿行,幽诡奇景与烟火市井交织生长,撑起一套独属于东方的志怪美学。 制作上,这部片系统性地试出了长篇叙事的 AIGC 全流程打法,逐步搭起一条覆盖创意开发、提示词工作流、视觉资产生成、AI 转绘、4K 输出到后期制作的完整技术链路。整部片子由一支20余人的团队完成,在导演、编剧这些影视核心工种之外,AI 资产效果师、提示词工程师这类新型创作者也深度入场,把 AI 技术与影视生产流程揉到了一处。 主创团队往《山海经》《子不语》《聊斋》等志怪典籍和民间传说里深挖,前后攒出200余个角色与视觉资产,搭起那个东方志怪的影像世界。影片 AI 导演刘海洋说,AIGC 那种快速迭代的劲头,让团队能不断试不同的视觉方案,在较短时间里跑通创意验证,让表达和画面持续打磨。 值得留意的是,这股 AI 拍片的势头并非孤例。爱奇艺今年4月还公布了《灵魂摆渡》首部全 AI 生成电影,定在2026年上映,编剧沿用原剧集的小吉祥天,由爱奇艺与长信传媒联手打造。爱奇艺创始人、CEO 龚宇此前曾发文表态,科技以人为本,永远为人服务,而不是为了取代人;AI 进影视圈,目的是服务观众、服务包括演员在内的所有创作者,把行业蛋糕做大,让创作者把精力留给创作本身。

View all 66 items →