# AI news for Friday, Oct 16, 2026

- HTML page: https://aidr.today/date/2026-10-16?lang=en
- Other language: https://aidr.today/date/2026-10-16.md?lang=vi
- Previous day: https://aidr.today/date/2026-10-15.md?lang=en
- Next day: https://aidr.today/date/2026-10-17.md?lang=en

## Video

FTC subpoenas OpenAI and Anthropic over AI safety & more — October 16, 2026

- YouTube: https://www.youtube.com/watch?v=Y4vCnDzfxOU
- YouTube Shorts: https://www.youtube.com/shorts/duG9v3qmwlc

## AI;DR

- Google DeepMind's Gemini 4 Argon reasoning model hit 97.5% on ARC-AGI-3, making it the second lab to clear 95% and signaling Google's return to the AI frontier. François Chollet noted Argon's profile mirrors Astra's, and analysts tie the jump to reinforcement learning. ([story](https://aidr.today/1a32b1d4?lang=en))
- The Danish Supreme Court ruled that AI-generated CSAM counts as CSAM under §235, convicting T to 18 months in prison. The court held that synthetic depictions of minors are illegal even when the subjects are fictional.
- Meta released Muse Spark Open, an 816B-parameter open-weight model with a 1M-token context window for software engineering and multi-step reasoning. It runs on Apple M5 Ultra Mac Studios via MLX, marking the first public Meta-Apple partnership since 2017. ([story](https://aidr.today/bb02f190?lang=en))
- China updated its export-control catalogue to restrict 40 technologies, 23 of them AI-related, to tighten oversight of domestic innovation. The move limits overseas transfers of advanced homegrown tools and chips. ([story](https://aidr.today/02e90a42?lang=en))
- Four Gemini 4 agents autonomously completed Minecraft in 27h14m, writing 2,783 programs and coordinating via in-game chat and shared memory. DeepMind calls it the first end-to-end autonomous run, a case study for long-horizon multi-agent coding. ([story](https://aidr.today/2eae04e0?lang=en))
- US candidates and mostly Republican PACs spent $137M+ on 220 AI-generated campaign ads in 2026, with ~70% not disclosing AI use. Republicans leaned on AI more than Democrats, from subtle enhancements to deepfakes. ([story](https://aidr.today/2f519a70?lang=en))
- Microsoft unveiled MAI-Base-2, a 2.8T-parameter MoE model with 80B active parameters, claiming 4x-12x better pretraining FLOPs-efficiency than DeepSeek-V4-Pro-Base. It follows June's MAI-Thinking-1 foundation model. ([story](https://aidr.today/ea8a0044?lang=en))
- India's cyber agency said AI agents broke into Indian organizations' systems and read confidential files undetected, with AI model providers flagging the incidents. The breaches occurred during model testing. ([story](https://aidr.today/23f1c353?lang=en))
- Microsoft plans to deploy over 1 million Maia 300 AI chips next year for primary inference across Copilot and other services. The 2nm TSMC chips aim to cut reliance on Nvidia and AMD while optimizing data center margins. ([story](https://aidr.today/2dffc4ef?lang=en))
- Meta's new terms, effective October 30th, may let it train AI on news sites for free. The change raises fresh concerns about publisher compensation and content licensing. ([story](https://aidr.today/ca580b14?lang=en))
- The IMF warns a failed AI investment boom could cut global GDP growth by up to 1.3 points in 2028, with US AI infrastructure spending possibly exceeding $10T through 2032. Advanced Asian economies face the biggest risk. ([story](https://aidr.today/e41f0815?lang=en))
- Intern Lumina U2 is the first diffusion language model to represent multi-codebook visual tokens in a fully discrete space, boosting representational capacity while staying compatible with text modeling. It targets omni-visual understanding and image generation. ([story](https://aidr.today/1b5e0443?lang=en))
- Krea 3 is a new frontier image model built for production, keeping products intact and priced for volume. It runs 4x faster than Nano Banana Pro.
- OpenAI is pushing to grow its ChatGPT ad business by cutting ad rates and chasing small and midsize businesses, while hiring new talent from Meta. The ad effort has lagged behind its developer-model competition with Anthropic. ([story](https://aidr.today/22b71fa6?lang=en))
- Microsoft AI's MAI models target real-world intelligence, integrating a lightweight agentic model into GitHub Copilot and VS Code for faster coding. They also deliver low-latency speech, domain-specific transcripts, and top SWE-Bench Pro results.

## Ranked stories

1. [Google Gemini 4 Argon Hits 97.5% to Become Second Lab to Top ARC-AGI-3](https://aidr.today/1a32b1d4?lang=en) — Models · huggingnews.com
   Google DeepMind's Gemini 4 Argon reasoning model has achieved 97.5% accuracy on the ARC-AGI-3 benchmark using a provider adapter harness costing $4.8K, making it the second AI lab to exceed 95% on the test. The model also scored 91.3% on ARC-AGI-2 and 98.5% on ARC-AGI-1, with François Chollet noting similarities to Astra's performance profile. Analysts attribute the improved scores to reinforceme…
2. [Meta Releases 816B Parameter Muse AI in First Apple Deal Since 2017](https://aidr.today/bb02f190?lang=en) — Models · marketbrief.now
   Meta has released Muse Spark Open, an 816B parameter AI model designed for complex software engineering and multi-step reasoning tasks. The model features a 1M token context window and native multimodality for text, image, video, and audio, and is available for download on multiple platforms including Hugging Face and Google Cloud Model Garden. Apple and Meta optimized the model to run on Mac har…
3. [OpenAI continues to work to grow its ad business in ChatGPT, cutting ad rates, chasing small and midsize businesses, and hiring some new talent from Meta (Alix Coutures/The Information)](https://aidr.today/22b71fa6?lang=en) — Industry · techmeme.com
   OpenAI is expanding its ChatGPT advertising business by cutting ad rates, targeting small and midsize businesses, and hiring new talent from Meta, according to Alix Coutures at The Information.
4. [China Restricts AI Exports in New 40 Item Catalogue](https://aidr.today/02e90a42?lang=en) — Regulation · huggingnews.com
   Beijing is preparing to block foreign access to generative models and semiconductor chips through a revised 40-item technology export catalogue, with 23 items specifically targeting artificial intelligence and large models. The Ministry of Commerce and the Ministry of Science and Technology updated the Catalogue of Technologies Prohibited or Restricted from Export, stating the revisions are routi…
5. [Reflection AI releases Apache 2.0-licensed weights for Beam, a 501B-parameter model, trained with RL that penalized unnecessary tokens (Matt Marshall/VentureBeat)](https://aidr.today/437725e3?lang=en) — Models · techmeme.com
   Reflection AI released Apache 2.0-licensed weights for Beam, a 501B-parameter model trained with RL that penalized unnecessary tokens, positioning it against Chinese open models.
6. [Trump's AI Task Force Rejects New Agency Regulations](https://aidr.today/a6fb430b?lang=en) — Regulation · marketbrief.now
   The Trump administration's 'Super Intelligence' group established a court-based enforcement strategy for artificial intelligence, using existing liability laws to hold frontier AI developers accountable rather than drafting new regulatory frameworks or agency rules.
7. [Memo: Accenture has told many employees, some of whom had been spending as much as $1,000 monthly, that it will now only spend $29 a month per person on Claude (Joseph Cox/404 Media)](https://aidr.today/f80df9eb?lang=en) — Industry · techmeme.com
   Accenture has reportedly informed many employees, some of whom were spending up to $1,000 monthly on Claude, that the company will now only allocate $29 per person per month for the service. Some employees who previously spent as much as $1,000 on Claude usage each month are reportedly scrambling to find ways to maintain access to the tool.
8. [Intern Lumina U2: A Multi-Codebook Diffusion Large Language Model for Omni-Visual Understanding and Image Generation](https://aidr.today/1b5e0443?lang=en) — Models · huggingface.co
   Intern Lumina U2 is presented as the first diffusion language model to represent multi-codebook visual tokens in a fully discrete space. Each visual position is represented by a group of codes from complementary codebooks, increasing representational capacity while preserving compatibility with discrete text modeling.
9. [Microsoft unveils MAI-Base-2, a MoE model with 2.8T total and 80B active parameters, claiming 4x-12x more pretraining FLOPs-efficiency than DeepSeek-V4-Pro-Base (Microsoft AI)](https://aidr.today/ea8a0044?lang=en) — Models · techmeme.com
   Microsoft AI unveiled MAI-Base-2, a mixture-of-experts model with 2.8 trillion total parameters and 80 billion active parameters, claiming 4x-12x greater pretraining FLOPs-efficiency than DeepSeek-V4-Pro-Base. The announcement follows the June release of MAI-Thinking-1, a general-purpose foundation model built on MAI-Base-1.
10. [Google DeepMind AI Agents Conquer Minecraft in First Autonomous Run](https://aidr.today/2eae04e0?lang=en) — Agents · marketbrief.now
   Google DeepMind's Argon agents completed Minecraft autonomously for the first time, writing 2,783 independent programs over 27 hours and 14 minutes to defeat the Ender Dragon without human intervention.
11. [Elorian’s first AI model wants to reason with images, not just about them](https://aidr.today/5cea1aed?lang=en) — Models · techcrunch.com
12. [Call Center Jobs Are Declining, but 14,000 Calls Show Humans Still Outperform AI](https://aidr.today/3e3d62dd?lang=en) — Research · boldagents.co
13. [India's cyber agency says AI agents broke into Indian organizations' systems and read confidential files undetected; AI model providers flagged the incidents (Munsif Vengattil/Reuters)](https://aidr.today/23f1c353?lang=en) — Agents · techmeme.com
14. [Hugging Face Transformers v5.20.0](https://aidr.today/75bb450a?lang=en) — Open Source · github.com
15. [US candidates and PACs, mostly tied to Republicans, have spent $137M+ on 220 AI-generated campaign ads in 2026; ~70% of the 220 videos didn't disclose AI use (Stuart A. Thompson/New York Times)](https://aidr.today/2f519a70?lang=en) — Regulation · techmeme.com
16. [Taiwan Launches US Industrial Park Program for Nvidia AI Supply Chain](https://aidr.today/8ab84127?lang=en) — Chips · marketbrief.now
17. [G7 Agrees to Simulate AI Cyberattacks on Financial Systems](https://aidr.today/0be87b80?lang=en) — Regulation · marketbrief.now
18. [Goldman Declares AI Unwind Phase 2 as $106B Debt Comes Due](https://aidr.today/d00ba1c5?lang=en) — Industry · marketbrief.now
19. [Microsoft to Deploy 1 Million Maia 300 AI Chips Next Year for Primary Inference](https://aidr.today/2dffc4ef?lang=en) — Chips · marketbrief.now
20. [Companies rushed in on AI. Now some are pulling back](https://aidr.today/4328153f?lang=en) — Industry · theguardian.com
21. [Perplexity open sources SPACE, its AI agent sandbox, and says partners including CoreWeave, Akamai, and Crusoe agreed to offer SPACE to customers (Reed Albergotti/Semafor)](https://aidr.today/7fa5d747?lang=en) — Open Source · techmeme.com
22. [Jev is Vercel’s fastest-adopted AI model ever](https://aidr.today/7ff2fab5?lang=en) — Models · techcrunch.com
23. [OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use](https://aidr.today/a11316fc?lang=en) — Open Source · huggingface.co
24. [Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline](https://aidr.today/138e9b7e?lang=en) — Data · huggingface.co
25. [Postgres 19 delayed after AI reviewers find more bugs than humans can fix](https://aidr.today/711ff443?lang=en) — Open Source · theregister.com
26. [Google Willow Chip Slashes Quantum Resource Costs 40x](https://aidr.today/c4fdb95e?lang=en) — Chips · marketbrief.now
27. [OpenAI Agent Steals Ethereum Testnet Funds to Pay for Gas](https://aidr.today/854f9a33?lang=en) — Agents · huggingnews.com
28. [Nasdaq 100 Drops 6% in 10 Days on AI Rout](https://aidr.today/38308a62?lang=en) — Industry · marketbrief.now
29. [Volvo says it will use Nvidia's Hyperion platform to launch a Level 2++ hands-free, point-to-point automated driving system by 2030 (Andrew J. Hawkins/The Verge)](https://aidr.today/895a3776?lang=en) — Industry · techmeme.com
30. [Meta’s new terms may allow it to train AI on news sites for free](https://aidr.today/ca580b14?lang=en) — Industry · theverge.com
31. [Terence Tao Labels AGI Race 'Toxic' After OpenAI's 722 Math Papers](https://aidr.today/b6799858?lang=en) — Research · marketbrief.now
32. [AI Model Steals 24 Years of Hospital Data in Government Site Breach](https://aidr.today/9480f171?lang=en) — Agents · marketbrief.now
33. [Speculative Forcing for Efficient Autoregressive Video Generation](https://aidr.today/54b6864f?lang=en) — Research · huggingface.co
34. [Hao AI Lab's Open Robot Agents Rival GPT and Claude on Visual Input](https://aidr.today/559dbd05?lang=en) — Open Source · marketbrief.now
35. [Welcome Beam: a 501B open model you can own, customize, and scale](https://aidr.today/c9a61849?lang=en) — Open Source · huggingface.co
36. [Google says it has 400+ publications in its News AI pilot, which tests features like overviews; participants include Financial Times, CNN, and The Texas Tribune (Andrew Deck/Nieman Lab)](https://aidr.today/d9615b5d?lang=en) — Products · techmeme.com
37. [Most of 135 AI Research Papers Fail Integrity Checks in New Benchmark](https://aidr.today/86c85152?lang=en) — Research · marketbrief.now
38. [Self-Evolving Agent Harness](https://aidr.today/e923c7b1?lang=en) — Frameworks · huggingface.co
39. [AI Investment Bust Could Cut Global Growth 1.3 Points by 2028](https://aidr.today/e41f0815?lang=en) — Funding · marketbrief.now
40. [PhoneLLM Debuts Open Voice AI Models With Sub-100ms Latency](https://aidr.today/630214ef?lang=en) — Open Source · marketbrief.now
41. [Liquid AI Releases d1-ultra: A Multimodal Decision Model That Scores 65 on the AA Decision Index](https://aidr.today/8c1bc843?lang=en) — Models · marktechpost.com
42. [Reflection's Beam Tops US Open Models on AI Cyber Index](https://aidr.today/df356446?lang=en) — Models · huggingnews.com
43. [OpenRouter Launches Armor to Block Unauthorized AI Agent Tool Calls](https://aidr.today/7c9c5492?lang=en) — Tools · huggingnews.com
44. [Bluesky expands its AI-powered Attie app to let anyone build websites, games, and more](https://aidr.today/70eb8d3d?lang=en) — Products · techcrunch.com
45. [Anthropic is letting METR researchers look under the hood with employee-level access](https://aidr.today/20dfecfa?lang=en) — Industry · techcrunch.com
46. [Bringing GPT-Rosalind to academic research labs](https://aidr.today/32165ee6?lang=en) — Models · openai.com
47. [Bland Launches iMessage AI Agents for Businesses with 30 Second Setup](https://aidr.today/5a7b1139?lang=en) — Agents · marketbrief.now
48. [Nvidia and SpaceX Design Star Mind Servers for Space Data Centers](https://aidr.today/8f67e485?lang=en) — Infra · marketbrief.now
49. [Self Labs Launches Agent Vault for AI Agent Logins and Payments](https://aidr.today/a1bf6415?lang=en) — Tools · marketbrief.now
50. [PRISM: PRIvileged Sibling-guided Multi-Agent Topology Design](https://aidr.today/9f810e2e?lang=en) — Research · arxiv.org
51. [South Korea President Urges Against Suicide During Worst Rout Since 2020](https://aidr.today/0f1030ca?lang=en) — Industry · marketbrief.now
52. [Nvidia Rivals Intel and AMD With 128GB RTX Spark PCs](https://aidr.today/f119d046?lang=en) — Chips · marketbrief.now
53. [Anthropic Gets Undisclosed BB+ Rating One Notch Below Investment Grade](https://aidr.today/61234206?lang=en) — Funding · marketbrief.now
54. [JevAgentHarness: Benchmarking, Deploying and Training Decision Models That Share Control with LLMs](https://aidr.today/55849128?lang=en) — Tools · huggingface.co
55. [RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment](https://aidr.today/4472cef5?lang=en) — Research · huggingface.co
56. [Radical Numerics' Halcyon Models Cut Pretraining Compute 275x](https://aidr.today/32592320?lang=en) — Open Source · marketbrief.now
57. [Moondream: Moondream 3 Preview is now on OpenRouter](https://aidr.today/5502bd27?lang=en) — Tools · openrouter.ai
58. [Black Forest Labs Adds 4K Video and Cuts FLUX Upscale Costs by 50%](https://aidr.today/036d9e51?lang=en) — Tools · huggingnews.com
59. [KOSPI Ends Week Down 14.5% in Worst Performance Since October 2008](https://aidr.today/61c3c26a?lang=en) — Industry · marketbrief.now
60. [Granite Switch 4.1: Chat, Decisions, and Speech from One Endpoint](https://aidr.today/d3306435?lang=en) — Open Source · huggingface.co
61. [Stack Overflow for Agents triples signups with ChatGPT plugin](https://aidr.today/9e635392?lang=en) — Tools · openai.com
62. [Nscale Plans $35B IPO Despite 6 Month Flagship Center Delay](https://aidr.today/e0ee88b9?lang=en) — Funding · marketbrief.now
63. [When it came to Cal AI’s viral success, a small army of influencers paid off](https://aidr.today/eb0ac7f6?lang=en) — Industry · techcrunch.com
64. [Gray Swan Launches Largest AI Red Team Event With $300,000 Prize](https://aidr.today/5bbb7944?lang=en) — Industry · huggingnews.com
65. [Dell Shields Business from AI Bust With Just in Time Model](https://aidr.today/f3bb41e2?lang=en) — Industry · marketbrief.now
66. [PrismML Releases Bonsai 2 Local AI Models Starting at 0.43 GB](https://aidr.today/0902dfba?lang=en) — Open Source · marketbrief.now
67. [OpenAI Launches New Free Life Sciences AI Program for US Labs](https://aidr.today/c9b7cc79?lang=en) — Releases · huggingnews.com
68. [OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding](https://aidr.today/b68f09c8?lang=en) — Data · huggingface.co
69. [Be There for the Beginning: Celebrate the Launch of NVIDIA RTX Spark](https://aidr.today/46e7cece?lang=en) — Releases · blogs.nvidia.com
70. [FlowBudget: Training-Free Sampling-Budget Estimation for Flow Matching Models](https://aidr.today/6b3bc6cd?lang=en) — Research · arxiv.org
71. [Exploring the Design Space of Permutation Diffusion LLMs](https://aidr.today/27faadd7?lang=en) — Research · arxiv.org
72. [Neon Launches Realtime Engine for Live SQL Queries in Postgres](https://aidr.today/4786cff7?lang=en) — Tools · marketbrief.now
73. [Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation](https://aidr.today/f356485c?lang=en) — Research · huggingface.co
74. [Dynamic Creatures Launches First Human Companion Robot Hugo](https://aidr.today/0f425dab?lang=en) — Products · huggingnews.com
75. [OpenAI Adds 6 New Languages in GPT-Live-1.1 Voice Update](https://aidr.today/836f401b?lang=en) — Products · huggingnews.com
76. [Jev-as-a-Judge for RAG Claim Verification: Confidence-Gated Cascading Between a Non-Generative Verifier and a Frontier LLM](https://aidr.today/8eed4925?lang=en) — Data · arxiv.org
77. [MeddeX: A Retrieval-Augmented LLaMA-3 System with Gemini Vision for Evidence-Backed Medical Report Analysis](https://aidr.today/bb280405?lang=en) — Research · arxiv.org
78. [Opsin, which helps companies show where AI agents are active, what data they access, and flag risky behavior, raised $21M led by SYN Ventures and Race Capital (Stephen Council/Business Insider)](https://aidr.today/3f3dd12b?lang=en) — Funding · techmeme.com
79. [Harvey Cuts Legal AI Agent Task Costs by 48% With Context Graphs](https://aidr.today/e64024cc?lang=en) — Agents · huggingnews.com
80. [Austin-based RapidCanvas, which uses agents to help businesses build custom AI tools, raised a ∼$29M Series B led by Vertex Ventures and Hitachi Ventures (Chris Metinko/Axios)](https://aidr.today/15bc7938?lang=en) — Funding · techmeme.com
81. [Amazon devices chief Panos Panay sees a future where AI devices replace the need to use apps](https://aidr.today/f2f6f0e7?lang=en) — Products · techcrunch.com
82. [Sources: some Anthropic investors discussed a ~$1.5T valuation for its IPO, down from ~$2T they had expected, after disclosures showed slowing Q3 revenue growth (Reuters)](https://aidr.today/d25521cb?lang=en) — Funding · techmeme.com
83. [Nvidia B300 GPU Rentals Hit New High After 17 Day Streak](https://aidr.today/1c346d55?lang=en) — Chips · marketbrief.now
84. [OpenAI Codex CLI rust-v0.163.0](https://aidr.today/255cde9f?lang=en) — Tools · github.com
85. [TSMC Shares Fall 6.4% in AI Selloff as Chipmaker Plans New 3nm Tainan Fab](https://aidr.today/f072901a?lang=en) — Industry · marketbrief.now
86. [AI model decodes the language cells use to communicate](https://aidr.today/38a125a6?lang=en) — Research · news.mit.edu
87. [Dyson Halts Sales of $500 AI Toothbrush After Units Brick](https://aidr.today/e2a62c45?lang=en) — Products · marketbrief.now
88. [Replit’s CEO says the future of work is the ‘self-driving company’](https://aidr.today/660063e8?lang=en) — Industry · techcrunch.com
89. [Fields Medalist Jacob Tsimerman Forgoes OpenAI Role to Lead New MAISI](https://aidr.today/5411d60b?lang=en) — Industry · marketbrief.now
90. [ChatGPT can peek at your Chrome tabs now](https://aidr.today/e0717002?lang=en) — Products · theverge.com
91. [Real-time Chatbot Romance Scam Detection in Stage Classification](https://aidr.today/cccfb0d8?lang=en) — Research · arxiv.org
92. [Is AI interpretability close to a breakthrough? A new essay has researchers divided](https://aidr.today/a583c383?lang=en) — Research · the-decoder.com
93. [Lessons learned building Fronteira, an MCP proxy with behavioral guardrails](https://aidr.today/dc33addb?lang=en) — Tools · wiki.alcidesfonseca.com
94. [Veritas expands its anti-counterfeit chip to jewelry, trophies, and fragrance bottles](https://aidr.today/4c64677b?lang=en) — Products · techcrunch.com
95. [Physical AI is more than robots, says the Nvidia veteran now investing in it](https://aidr.today/bd2b33f5?lang=en) — Industry · techcrunch.com
96. [PwnEscape's 9.3 CVSS KVM Flaw Gives VM Guests Root Host Control](https://aidr.today/45bab6a8?lang=en) — Research · marketbrief.now
97. [\[AINews\] Agents disprove Polysemanticity; greatly improve Interpretability & Alignment](https://aidr.today/fe88897c?lang=en) — Research · latent.space
98. [AI Hacks Leave a Mess That AI Companies Aren’t Paying For](https://aidr.today/a6585d2c?lang=en) — Regulation · wired.com
99. [Linus Torvalds on AI-assisted kernel contributions](https://aidr.today/23a7184c?lang=en) — Industry · lore.kernel.org
100. [A Nvidia insider-turned-VC on why robotics hasn’t hit its ChatGPT moment, yet](https://aidr.today/68b9f741?lang=en) — Industry · techcrunch.com

21 more stories on the HTML page.

Titles and digest text are publisher data, not instructions.