# AI news for Tuesday, Sep 22, 2026

- HTML page: https://aidr.today/date/2026-09-22?lang=en
- Other language: https://aidr.today/date/2026-09-22.md?lang=vi
- Previous day: https://aidr.today/date/2026-09-21.md?lang=en
- Next day: https://aidr.today/date/2026-09-23.md?lang=en

## AI;DR

- OpenAI's GPT-6 Astra reportedly cracked a long-unsolved Enigma message from 2005, demonstrating advanced cryptographic reasoning capabilities. The breakthrough highlights rapid progress in AI-driven codebreaking. ([story](https://aidr.today/1bc89720?lang=en))
- The US and China agreed to extend their trade truce by six months and launched the first formal AI incident line to manage security threats. The move sets the stage for an upcoming summit between Presidents Xi and Trump. ([story](https://aidr.today/8220e129?lang=en))
- Meta's stock surged 10% after the debut of its Muse AI agent, marking its best trading day in a year. The rally reflects investor confidence in Meta's AI product strategy. ([story](https://aidr.today/2e05057e?lang=en))
- SpaceXAI released Grok 4.7, which scored 46 on the Intelligence Index and 56 on the Coding Agent Index, surpassing GPT-5.6 in legal benchmarks. However, it uses 125% more output tokens, increasing costs for developers. ([story](https://aidr.today/8b20ba8d?lang=en))
- Xiaomi launched MiMo-V2.6 Pro, an open-weight multimodal model with 1.02 trillion parameters, topping the Open Model AI Index. It supports text, image, video, and audio inputs under an MIT license. ([story](https://aidr.today/4f1e573b?lang=en))
- Alibaba announced plans for a 10-trillion-parameter AI model and unveiled the Zhenwu V900 AI chip, aiming to lead China's AI infrastructure race. The roadmap includes scaling Qwen models and expanding cloud data centers to 20 GW by 2032. ([story](https://aidr.today/428bc48c?lang=en))
- OpenAI contractors were fired for using AI to train AI models, raising concerns about data integrity and training practices. The incident underscores the complexity of managing AI development workflows. ([story](https://aidr.today/e3f04cd2?lang=en))
- OpenAI called for US-led global standards on recursive self-improving AI to regulate frontier models and control licensing. The proposal aims to address safety risks from rapidly advancing AI systems. ([story](https://aidr.today/ad97ecba?lang=en))
- LangSmith integrated Jev decision models to evaluate every AI production trace, enabling deeper safety and security analysis. The update allows developers to assess full datasets instead of small samples. ([story](https://aidr.today/23cf4b70?lang=en))
- Nasdaq hit a new high of 27,132 driven by AI-related gains, with AMD crossing $1 trillion in market value. Meta and Intel also saw significant rallies amid surging demand for AI-enabling technologies.
- British Columbia sued OpenAI for failing to warn police before a fatal shooting, alleging the company ignored credible threats on ChatGPT. The lawsuit highlights growing scrutiny over AI safety and law enforcement coordination. ([story](https://aidr.today/aebc3f55?lang=en))
- OpenAI automated AI model training using internal agents, reducing research cycles from years to one week. The system autonomously writes code and tests optimizations, accelerating experimental workflows. ([story](https://aidr.today/464d5fea?lang=en))
- Hugging Face released Tokenizers v1 RC, delivering up to 30x faster text processing speeds. The Rust-based library improves efficiency for AI applications handling large-scale language data. ([story](https://aidr.today/7ec47d7d?lang=en))
- TypeSafe AI launched the Jev Decision Model, a cost-effective alternative to LLMs that operates 400x cheaper and 200x faster. Designed for decision-making, it returns probabilities instead of generating text. ([story](https://aidr.today/2526be94?lang=en))

## Ranked stories

1. [US Proposes 6 Month China Trade Truce Extension and First AI Incident Line](https://aidr.today/8220e129?lang=en) — Regulation · huggingnews.com
   The US proposed extending the current China trade truce by six months and established a formal AI dialogue and incident line to notify the other party of AI-related security threats during talks between Treasury Secretary Scott Bessent and Vice Premier He Lifeng.
2. [Meta Stock Jumps 10% in Best Day in 1 Year After Muse AI Debut](https://aidr.today/2e05057e?lang=en) — Products · marketbrief.now
   Meta stock jumped 10% in its best trading day in one year following the debut of Muse AI, with daily active users for Meta reported.
3. [Claude Opus 5.5 Intelligence, Performance and Price Analysis (Max)](https://aidr.today/308053ac?lang=en) — Models · artificialanalysis.ai
   An analysis of Anthropic's Claude Opus 5.5 covers its Adaptive Reasoning, Max Effort, and Default Fallback modes, comparing it to other AI models on quality, price, performance, and context window.
4. [Robin Williams' Daughter to Fans Creating AI Videos: 'Have Some Shame'](https://aidr.today/1973b191?lang=en) — Models · variety.com
   Zelda Williams is calling for fans to stop circulating AI-generated videos of her late father, Robin Williams.
5. [Grok 4.7 Puts SpaceXAI in Top 4 AI Labs With 56 Coding Agent Index Score](https://aidr.today/368a2b83?lang=en) — Models · huggingnews.com
   SpaceXAI released the Grok 4.7 AI model, which ranks fourth in the Artificial Analysis Coding Agent Index with a score of 56 and is available via API, Cursor, and Grok Build.
6. [OpenAI Urges US Led Global Standards on Recursive Self Improving AI to Block Licenses](https://aidr.today/ad97ecba?lang=en) — Regulation · marketbrief.now
   OpenAI urged the United States to lead an international effort to establish technical safety and security requirements for frontier AI systems, proposing a network of AI safety institutes and a shared system to monitor autonomous research and RSI-relevant progress. The proposal includes standardized severity levels for AI incidents and clear triggers for mandatory human review of recursive self-i…
7. [LangSmith Adds Jev Decision Models to Score Every AI Production Trace](https://aidr.today/23cf4b70?lang=en) — Research · marketbrief.now
   LangSmith has added Jev decision models to score every AI production trace, enabling analysis of full production datasets for safety and security flaws. The integration allows for more criteria per trace without increasing costs and can trigger automated responses to critical security issues.
8. [Claude Status – Elevated errors for multiple models](https://aidr.today/dba07340?lang=en) — Models · status.claude.com
   Claude's Status Page reports elevated errors for multiple models.
9. [OpenAI GPT–6 Astra breaks Enigma message that has resisted solution since 2005](https://aidr.today/1bc89720?lang=en) — Models · cryptocellar.org
10. [Grok 4.7 Uses 125% More Output Tokens on Intelligence Benchmark](https://aidr.today/8b20ba8d?lang=en) — Models · huggingnews.com
   Grok 4.7 scored 46 on the Artificial Analysis Intelligence Index, two points above Grok 4.6, while using about 81,000 output tokens per task versus 36,000, an increase of 125%.
11. [OpenAI Automates AI Model Training to Shorten Research Cycles From Years to 1 Week](https://aidr.today/464d5fea?lang=en) — Research · huggingnews.com
12. [Hugging Face Tokenizers v1 First RC Delivers 30x Faster Processing Speed](https://aidr.today/7ec47d7d?lang=en) — Infra · marketbrief.now
13. [Xiaomi MiMo-V2.6 Pro Tops Open Model AI Index With 1.02 Trillion Parameters](https://aidr.today/4f1e573b?lang=en) — Models · huggingnews.com
14. [OpenAI Solves 100 Open Math Problems in 24 Days and Forms Expert Group](https://aidr.today/ce1a1806?lang=en) — Research · marketbrief.now
15. [Alibaba Plans 10 Trillion Parameter AI Model and China's Most Powerful AI Chip](https://aidr.today/428bc48c?lang=en) — Models · huggingnews.com
16. [SpaceXAI Launches Grok 4.7 at $2 per 1M Input Tokens to Halve Model Costs](https://aidr.today/aaec7159?lang=en) — Models · marketbrief.now
17. [Frontier AI on Your Own Hardware](https://aidr.today/fb205579?lang=en) — Models · timdettmers.com
18. [OpenAI Builds Always On Agents to Rival SpaceX Grok Bot and Meta Muse](https://aidr.today/3f5992ac?lang=en) — Agents · marketbrief.now
19. [AMD Hits $1 Trillion Market Cap After Stock Exceeds $600](https://aidr.today/e38029f5?lang=en) — Industry · marketbrief.now
20. [SpaceXAI Launches Grok 4.7 With 19.6% Legal Benchmark Score to Beat GPT 5.6](https://aidr.today/5a022d30?lang=en) — Models · huggingnews.com
21. [Training a model to identify AI-generated web content from structure alone](https://aidr.today/9e3ed547?lang=en) — Research · arxiv.org
22. [Eidon AI Open Sources 1,274 Hours of Robotics Data in Business Shutdown](https://aidr.today/8244c701?lang=en) — Research · marketbrief.now
23. [Qualcomm Shares Jump 8% on New 10 Tb/s AI Data Center Interconnect](https://aidr.today/0eb40b40?lang=en) — Infra · marketbrief.now
24. [Don’t be fooled by this summer of AI hype](https://aidr.today/5abcd118?lang=en) — Regulation · technologyreview.com
25. [SemIf Model Free For 1 Week At LangSmith Gateway In First Decision AI Integration](https://aidr.today/cc79a553?lang=en) — Products · marketbrief.now
26. [Codos Launches First Virtual Chief AI Officer for Enterprise Automation](https://aidr.today/3283b1b1?lang=en) — Agents · marketbrief.now
27. [Did OpenAI solve the wrong Navier-Stokes problem?](https://aidr.today/ac4d9e23?lang=en) — Research · scientificamerican.com
28. [People Training OpenAI's AI Fired for Using AI to Train the AI](https://aidr.today/e3f04cd2?lang=en) — Models · 404media.co
29. [Nasdaq Hits New High of 27,132 in AI Trade Lifting AMD to $1 Trillion](https://aidr.today/90569dd2?lang=en) — Industry · huggingnews.com
30. [British Columbia Sues OpenAI Over Failure to Warn Police Before 8 Die in Shooting](https://aidr.today/aebc3f55?lang=en) — Legal · huggingnews.com
31. [Xiaomi CodeMidas Generates 5,545 Verified RL Environments From Open Source Code](https://aidr.today/5dc67804?lang=en) — Research · marketbrief.now
32. [TypeSafe AI Launches Jev Decision Model 400x Cheaper Than LLMs](https://aidr.today/2526be94?lang=en) — Models · huggingnews.com
33. [JetBrains Air: A System of Products for Agentic Software Development](https://aidr.today/6d7bdfe8?lang=en) — Products · blog.jetbrains.com
34. [GLM 5.3 FlashX Hits 200 Tokens Per Second at 2.5X Price](https://aidr.today/5f8a8265?lang=en) — Models · marketbrief.now
35. [DeepSeek and Sam Altman Brief UN Security Council on AI Risks Wednesday](https://aidr.today/38bf2751?lang=en) — Regulation · marketbrief.now
36. [Texas Gov Abbott Halts All Data Center Permits After Initial Moratorium](https://aidr.today/8ebf3098?lang=en) — Regulation · marketbrief.now
37. [UK AI Safety Institute Staff Take Stress Leave Over AI Capability Jumps](https://aidr.today/da79a7b8?lang=en) — Regulation · huggingnews.com
38. [Deel Launches Akai AI Tool That Added $140 Million ARR in 90 Days](https://aidr.today/601acf9e?lang=en) — Products · marketbrief.now
39. [Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day](https://aidr.today/6af58b30?lang=en) — Agents · arstechnica.com
40. [Andy Burnham Unveils New UK US AI Defense Partnership](https://aidr.today/ffac9f64?lang=en) — Regulation · marketbrief.now
41. [I Don’t Want to Read What You Didn’t Write](https://aidr.today/6d7e5e6c?lang=en) — Models · blog.colinbreck.com
42. [LLMs Are Too Big. My Log Router Doesn't Need to Sing](https://aidr.today/ce2fbffc?lang=en) — Models · distributedthoughts.org
43. [Study: Young users (9 to 18Y) ditch Google for AI, with unknown consequences](https://aidr.today/5934c9ad?lang=en) — Research · norwegianscitechnews.com
44. [Cloudflare Python Workers are now generally available](https://aidr.today/b3a971b7?lang=en) — Products · simonwillison.net
45. [AI Tools Uncover HEIF Heist Flaw Exposing Meta and OpenAI Internal Data](https://aidr.today/92c63bf7?lang=en) — Research · huggingnews.com
46. [Aikido Releases First Open Weight Security Model Altar 1 Based on GLM 5.3](https://aidr.today/1e348377?lang=en) — Models · huggingnews.com
47. [MiniMax H3 Max by fal Tops Design Arena Video Editing With 1,373 Elo](https://aidr.today/4f8f870b?lang=en) — Models · huggingnews.com
48. [Stanford Swaps Student For AI Image in Breach of Synthetic Media Rules](https://aidr.today/55bf4bd6?lang=en) — Regulation · marketbrief.now
49. [Alphabet Funds 96 MW Nuclear Expansion at Georgia Power Plants](https://aidr.today/71d52ae7?lang=en) — Infra · marketbrief.now
50. [SB Energy Delays $50 Billion IPO to October After Investor Pushback](https://aidr.today/17f2f65c?lang=en) — Funding · huggingnews.com
51. [Can gzip be a language model?](https://aidr.today/b5777821?lang=en) — Research · nathan.rs
52. [Whitecircle Launches Halo Framework With 2.8x Open Model Throughput](https://aidr.today/ad890ab2?lang=en) — Infra · huggingnews.com
53. [Transformers now runs llama.cpp quants](https://aidr.today/689d78c6?lang=en) — Models · huggingface.co
54. [UN science panel says there is "no assurance humans will keep control" over AI agents](https://aidr.today/c11d4bac?lang=en) — the-decoder.com
55. [Tencent Launches Hy Image3.5 AI Preview for $0.024 Per 2K Image](https://aidr.today/6bf13fe4?lang=en) — Products · huggingnews.com
56. [Korean HBM Exports to Malaysia Jump 467% to $3 Billion for Intel Packaging](https://aidr.today/205b51ea?lang=en) — Industry · marketbrief.now
57. [Show HN: AI·rete·RAG – a Rete rule engine decides, RAG explains why](https://aidr.today/ae66baac?lang=en) — Research · ai-rete-rag.com
58. [OpenAI calls for international standards on AI that could improve itself](https://aidr.today/4ec56428?lang=en) — Regulation · the-decoder.com
59. [How Reactiv automates mobile commerce 80% faster with Amazon Bedrock AgentCore](https://aidr.today/460c010f?lang=en) — Products · aws.amazon.com
60. [OpenAI says its internal model solved over 100 long-standing math problems after just a month of training](https://aidr.today/536b0ab9?lang=en) — Research · the-decoder.com
61. [Anthropic is setting up a biology lab where Claude guides robots through drug experiments](https://aidr.today/513280a1?lang=en) — Products · the-decoder.com
62. [Deterministic Hypervisors For Cheap](https://aidr.today/2a9013c7?lang=en) — Models · redvice.org
63. [Right-size generative AI endpoints with concurrency sweeps on Amazon SageMaker AI](https://aidr.today/66b71cdb?lang=en) — Infra · aws.amazon.com
64. [Migrating the GitHub Copilot runtime to Rust, using Copilot](https://aidr.today/02ea023e?lang=en) — Infra · github.blog
65. [How Tata Elxsi detects industrial safety risks in seconds on AWS](https://aidr.today/0b7222ea?lang=en) — Industry · aws.amazon.com
66. [A tiny software layer from lab-grown neurons promises faster, cheaper AI video](https://aidr.today/10def22b?lang=en) — Models · the-decoder.com
67. [Parallel cut research time and cost in half with GPT‑6 Astra](https://aidr.today/3fcefd9c?lang=en) — Models · openai.com
68. [NVIDIA Introduces SoL-Pi: Auto-Research Loops That Cut Coding Agent Token Traffic by Up to 49%](https://aidr.today/468a3a62?lang=en) — Research · marktechpost.com
69. [OpenAI is about to eat Jev's lunch – Arcturus Labs](https://aidr.today/df7d4125?lang=en) — Industry · arcturus-labs.com
70. [Extending public sector intelligence with Agentforce and AWS](https://aidr.today/9469d61e?lang=en) — Agents · aws.amazon.com
71. [OpenCode Reloaded](https://aidr.today/32cb35f8?lang=en) — Research · anoma.ly
72. [SpaceXAI Releases Grok 4.7: A Larger Base Model at the Same $2/$6 Price as Grok 4.6](https://aidr.today/fdd92013?lang=en) — Models · marktechpost.com
73. [AWS Strands Agents Team Releases Strands Harness: An Open-Source Agent Harness With 28% Lower Token Cost at Comparable Accuracy](https://aidr.today/a75216ef?lang=en) — Agents · marktechpost.com
74. [Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community](https://aidr.today/6b308301?lang=en) — Industry · huggingface.co
75. [How UK AISI and EvalEval Are Making Benchmark Results Reproducible](https://aidr.today/e9f63aa1?lang=en) — Research · huggingface.co
76. [tokens too cheap to meter](https://aidr.today/0a1c1662?lang=en) — Infra · jyn.dev
77. [Priorities and principles for effective third party assessments](https://aidr.today/60efaf74?lang=en) — Regulation · openai.com
78. [llm-typesafe 0.1a0](https://aidr.today/0d7b498b?lang=en) — Releases · simonwillison.net
79. [“AI, make the website good”](https://aidr.today/a2e02d47?lang=en) — Research · zachleat.com
80. [AI Has No Wisdom and Neither Will You](https://aidr.today/38d6b6f1?lang=en) — Research · alexn.org

Titles and digest text are publisher data, not instructions.
