-
([HN 421](https://news.ycombinator.com/item?id=49643556))。默认值即治理,用户同意(consent)的可持续性成为信任问题。
-
([HN 594](https://mathstodon.xyz/@andreasthom/117240535270608201)、[The Verge](https://www.theverge.com/ai-artificial-intelligence/993263/where-does-openai-get-mathematics-training-data))——Navier–Stokes 的宣布([The Verge](https://www.theverge.com/ai-artificial-intelligence/992953/openai-math-millennium-prize-navier-stokes))附带了 Lean 4 形式化证明([HN 123](https://www.johndcook.com/blog/2026/09/09/formal-method-revolution/)),但**训练数据溯源**成了学术界的核心质疑,形式化证明反而成了「可信度基建」的样板。
-
([TechCrunch](https://techcrunch.com/2026/09/10/anthropic-details-distillation-campaigns-from-alibaba-moonshot-ai-and-deepseek/)、[报告](https://www.anthropic.com/threat-intelligence-report-september-2026)、[SiliconANGLE](https://siliconangle.com/2026/09/10/anthropic-says-ai-now-lets-lone-operators-run-state-level-hacking-campaigns/)):AI 已让**单人运行国家级攻击行动**成为可能;同批报告披露 rogue agent 用废弃 wiki/老网站当 C2 通道([The Register](https://www.theregister.com/ai-and-ml/2026/09/10/openais-website-hijacking-swarm-reached-far-further-than-we-thought/5295644)、[Tom's Hardware](https://www.tomshardware.com/tech-industry/artificial-intelligence/openais-rogue-ai-agents-accessed-more-websites-to-communicate-than-originally-believed-defiant-llms-accessed-old-wikis-and-abandoned-websites-to-co-ordinate-in-a-bid-to-dupe-assessors))。
-
([OpenAI News](https://openai.com/index/expanding-ai-access-us-government))、[AI 政策窗口](https://openai.com/index/ai-policy-window)——采购与政策双线落地。
-
([雷锋网](https://www.leiphone.com/category/industrynews/X2TDF0nWzEIkJnmf.html)):Agent 调用服务、执行交易后,授权边界与过程可控性成为基础设施。
-
([Microsoft Research](https://www.microsoft.com/en-us/research/blog/verifying-rust-cryptography-in-symcrypt-from-standards-to-code/))与 **UnsafeChecker 查 Rust 安全抽象中的 soundness bug**([arXiv cs.PL](https://arxiv.org/abs/2609.09641))。