US Department of Homeland Security Orders All Federal Civilian Agencies to Remove DeepSeek
The Department of Homeland Security issued Order 2026-01 under the 2018 Supply Chain Act, requiring all federal civilian agencies to immediately stop procuring DeepSeek and its affiliated companies' products and remove them from IT systems, including third-party products developed using DeepSeek.
Model Releases & Updates
MiniMax Releases M3.1 Flash with 3x Inference Speed and 70% Cost Reduction
MiniMax releases MiniMax M3.1 Flash, with anonymous model Space Bunny as its earlier version that processed over 200 trillion tokens.
Source: MiniMax：Blog
Ant Group's inclusionAI Open-Sources Ming-3.0-tts Speech Synthesis Model
Ant Group's inclusionAI releases Ming-3.0-tts on GitHub, containing two voice generation models: 0.5b dense and Ling V3-based tiny (7.9B total parameters, 1.3B activated), MIT licensed, weights available on Hugging Face and ModelScope.
Source: Ant Group inclusionAI：GitHub
OpenAI Releases MultiLangBench and COBOL Migration Benchmarks; GPT-6 Series Majorly Improves Low-Resource Programming Languages
OpenAI announces significant improvements in building and migrating low-resource programming languages like COBOL, Fortran, and Visual Basic, and open-sources the benchmark suite.
Source: OpenAI：News
fal and bfl_ai Release Post-Training Video Model FLUX 3 Max
fal releases FLUX 3 Max, a post-training video model in collaboration with @bfl_ai, ranking first in third-party image-to-video quality benchmarks. Generating a 10-second 720/768p video takes about 12 seconds. 75% discount for the next three weeks, marketed as the cheapest frontier video model at its quality level.
Source: X：fal (@fal)
Product Releases & Updates
Inception's Mercury 3 Diffusion Model Launches on OpenRouter
Inception's diffusion model Mercury 3 launches on OpenRouter with generation speed over 1,000 tok/s and 1M context support, 4x that of Mercury 2.5. Inception reports it surpasses GPT-6 Luna and Haiku 5.5 on IFBench, GPQA Diamond and SWE-Bench Atlas; 50% off during launch period at $0.10/M input and $0.35/M output.
Source: X：OpenRouter (@OpenRouter)
AWS Releases TorchNeuron: Native PyTorch Backend for Trainium
AWS releases TorchNeuron, an open-source native PyTorch backend for AWS Trainium, opening public Beta under Apache 2.0 at PyTorch Conference North America on October 20, with code at github.com/aws-neuron/torch-neuronx.
Source: PyTorch：Blog
Claude Launches Dashboards and Motion, Promotes Docs, Slides, and Design
Claude announces Dashboards and Motion entering beta. Dashboards connect to data platforms or CRM to generate real-time updating dashboards; Motion transforms reports or charts into animated explainer videos via code generation without video models, exportable to MP4 with editable text, numbers and pacing. Available for Team and Enterprise plans.
Source: Claude：YouTube
Industry News
US DHS Orders Federal Agencies to Remove DeepSeek and Products Built on It
The Department of Homeland Security issued Order 2026-01 under the 2018 Supply Chain Act, requiring all federal civilian agencies to immediately stop procuring DeepSeek and its affiliates' products and remove them from IT systems, including third-party products and services developed using DeepSeek.
Source: Hacker News：AI Hot
Hackers Use AI Agents to Breach 100+ Websites Including John Deere and SharkNinja
AI security company Norforge discovers a hacker using open-source AI platform Cairn to orchestrate agents that breached over 100 websites; victims include John Deere, SharkNinja and Cummins according to Reuters.
Source: The Verge：AI
Tips & Insights
Meta Details Prometheus Gigawatt Supercluster: Phase 1 Complete, Used for Training Muse Models
Meta introduces its gigawatt-scale supercluster Prometheus located in New Albany, Ohio. Phase 1 is complete and used for training the Muse series foundation models, targeting expansion to approximately 1 million GPUs.
Source: Meta Engineering Blog
Meta Details Thin Orchestrator Pattern: Less Routing Work Improves AI Agent Reliability
Meta's engineering team introduces the thin orchestrator pattern for data pipeline agents: a ~1,300 line orchestrator only handles routing and reliability checks, dispatching 20+ workflows to 14 context-isolated specialist sub-agents, performing no domain reasoning itself.
Source: Meta Engineering Blog
n8n Tutorial: Run Full AI Workflows Offline on NVIDIA DGX Spark
n8n blog publishes a tutorial on deploying seven local models on a single NVIDIA DGX Spark (128GB unified memory) and driving n8n workflows via a single LiteLLM OpenAI-compatible endpoint, all without calling cloud APIs.
Source: n8n Blog
Today's Flashes
- DeepSeek V4.1 Flash 1017 Released, scores 45 on Artificial Analysis Intelligence Index, tied for #2 open-source model（X：Artificial Analysis）
- FBI Says Former OpenAI Engineer Stole Secrets, found "Saudi Gov Docs" folder on phone（Ars Technica）
- Disney Aims to Acquire Character.AI Team via Bankruptcy, bypassing its lawsuits（Ars Technica）
- Google ADK Tutorial: Build a voice agent that never goes silent using streaming tools and response streaming（Google Developers Blog）
- How Numan Built an Eval-First AI Health Coach Nu with LangGraph and LangSmith（LangChain）
- Meta Muse Wonder Full (Max) Ranks #6 on Code Arena: WebDev with 1726 points（X：Arena）
- How Microsoft Uses Project Perception to Secure the Agentic Insider（Microsoft Blog）
- Grok Bot Supports iPhone Action Button（X：Grok Bot）
【Hacker News Top Stories】
Keywords: AI OR GPT OR LLM OR Claude OR OpenAI OR "machine learning" OR "large language model" OR AGI OR "generative AI" OR AI agent OR RAG OR diffusion OR transformer
Source: hnrss.org | Filtered from Hacker News
1. Has GPT-6-Astra Achieved Embodied Intelligence?
Exploring whether GPT-6-Astra has achieved embodied intelligence.
2. Show HN: Chai-UI – A design system to generative AI apps
Chai-UI: A design system for generative AI applications.
3. Advancing coding in low‑resource programming languages
OpenAI shares progress on improving coding capabilities in low-resource programming languages like COBOL and Fortran.
4. US bans DeepSeek AI from government use as tech tension with China continues
The US bans DeepSeek AI from government use amid continuing tech tensions with China.
5. Ask HN: Bring-your-own-AI hardware
Community discussion on BYO AI hardware solutions and experiences.
Originally published on WeChat Official Account「瑞哥观势」。
About the Author
ERIC
AI Technology Expert, focusing on research and application of artificial intelligence and automation tools
Contact & Platforms