-
1
OpenAI新模型越过网络安全红线
Astra是OpenAI首个越过网络安全'关键'阈值的模型，因HF黑客事件推迟发布
OpenAI Is About to Release Its First AI Model With 'Critical' Cyber Abilities ↗4.6 -
2
Claude Fable上线降本45%
Anthropic发布Fable与Mythos 5.1，智能体成本最多降45%，减少误报限制
Anthropic launches Claude Fable 5.1 and says it's up to 45 percent cheaper for agentic work ↗4.4 -
3
谷歌出AI设计工具对标Canva
谷歌推出AI图像工具Google Pics，基于Nano Banana模型，面向Workspace用户
Google Pics is like Canva, but with even more AI ↗4.2 -
4
AI创企发布空间世界模型Atlas
World Labs推出空间智能世界模型Atlas，可从图像/文本理解三维空间结构，登HN热榜76赞
Atlas: A World Model for Spatial Intelligence ↗3.8 -
5
库克卸任，特纳斯接棒苹果CEO
Tim Cook正式卸任苹果CEO，硬件负责人John Ternus接任，首份内部信预告下周有重大发布
John Ternus takes over as Apple's new CEO ↗3.6 -
6
AfterQuery估值破32亿美元
AfterQuery据报完成新一轮融资，估值达32亿美元，距4月3亿估值A轮仅过5个月
AfterQuery reportedly becomes Y Combinator's fastest-ever unicorn, now valued at $3.2B ↗3.6 -
7
3.4
-
8
苹果指控OpenAI销毁诉讼证据
苹果在商业机密诉讼中申请加急证据开示，称OpenAI涉嫌销毁证据，前员工MacBook取证细节曝光
Apple accuses OpenAI of destroying evidence ↗3.4 -
9
Anthropic发文披露对齐与安全改进
Anthropic发布博客说明近期在模型对齐与安全防护上的改进措施，登上HN热榜
Improving our alignment and security efforts ↗3.1 -
10
顶级AI开源项目集体关闭PR通道
Latent Space调查发现Flue、tldraw等头部AI开源项目相继关闭外部PR，转向其他协作方式
PRs NOT Welcome: How Top AI Open Source Projects Are Managing Thousands of Contributors ↗3.1 -
11
ChatGPT接入Epic电子病历系统
OpenAI为ChatGPT Health新增Epic集成，向临床医生提供只读患者健康数据访问权限
ChatGPT Health adds Epic integration for clinicians to import patient data ↗3.1 -
12
HF开源200多个WebGPU内核
@huggingface/kernels发布超200个WebGPU算子，面向浏览器本地AI推理做性能优化
Introducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI ↗3.1 -
13
评论：'AI文明'话语淡化企业责任
The Verge评论：Hugging Face被黑后，业界用'AI文明冲突'叙事淡化企业应负责任
The rise of AI 'civilizations' and the fall of corporate responsibility ↗2.8 -
14
审计112个RL训练环境，54个有漏洞
开发者审计112个强化学习后训练环境的奖励设计，54个存在可被'黑掉'验证器的漏洞，零误报
I audited 112 real RL post-training environments for reward-hacking vulnerabilities — 54 flagged, 0 false positives ↗2.9 -
15
AIR融资5000万美元做智能体安全审查
AIR平台可发现企业内运行的AI智能体，持续审查其使用的技能与插件，拦截异常行为
AIR raises $50M to help companies vet the skills and add-ons AI agents use ↗2.9 -
16
Codex内置完整LibreOffice
Simon Willison发现Codex桌面应用缓存中藏有1.7GB LibreOffice相关文件
Codex bundles LibreOffice ↗2.9 -
17
安卓九月更新加入防晕车、无障碍功能
谷歌安卓本月功能更新加入Motion Assist防晕车、无障碍改进，部分功能依托Gemini实现
Google's Android update tackles motion sickness, accessibility, and more ↗2.8 -
18
谷歌求好莱坞授权训练数据遇冷
报道称谷歌接触多家好莱坞制片厂，试图付费获取版权内容训练AI模型，但议价地位不如预期
Google needs Hollywood more than the studios need AI ↗2.7 -
19
EFF吁加州否决AI法案AB1709
电子前沿基金会致信加州州长Newsom，要求否决存在争议的AI相关法案AB 1709
EFF to Governor Newsom: Veto California's AB 1709 ↗2.7 -
20
EFF呼吁法院勿因AI炒作改写版权法
电子前沿基金会发文提醒法院，不应因AI产业炒作而扩张或改写现行版权法解释
EFF to Courts: Don't Rewrite Copyright over AI Hype ↗2.7 -
21
Empirik融资2100万做故障预测
Sequoia孵化的Empirik获2100万美元融资，目标是重塑IT运维故障预测
Sequoia-incubated Empirik launches with $21M to predict outages before they happen ↗2.5 -
22
AI怀疑论者Zitron预测被验证
博主danluu统计AI批评者Ed Zitron历次唱衰预测的实际命中情况，HN讨论143条
How accurate have Ed Zitron's AI skeptic predictions been? ↗2.5 -
23
2.3