Sponsored by
Today’s Highlights:
🤖 US Bans New Chinese Humanoid Robots
🧮 OpenAI’s Astra Cracks 10 Major Math Problems
🚀 Alibaba Unveils Its Biggest AI Model
🌍 Google Pauses AI Image Generation in Earth
🎥 MiniMax Launches Open-Weight 2K Video Model
🗂️ Perplexity Renames Spaces to Projects
⚖️ German Court Rules Against Suno in Copyright Case
Read Time: 4 mintues
LATEST AI NEWS
HUMANOID ROBOTS
Image Source: Getty Images
The US has effectively blocked new foreign-produced humanoid and quadruped robots from entering its market, with the restrictions expected to hit Chinese manufacturers the hardest.
The Details:
The Federal Communications Commission added foreign-produced advanced robotic devices to its Covered List, preventing new models from receiving the authorization required for import, marketing, or sale in the US.
The rule covers mobile robots such as humanoids and four-legged robot dogs, preventing new models from receiving the FCC authorization required for import, marketing, and sale.
The restrictions apply only to new and unreleased models. Previously approved robots can still be used, imported, and sold in the US.
Existing approved devices can continue receiving security, compatibility, software, and firmware updates until at least January 1, 2029. Manufacturers can also seek conditional approval from the Department of War.
China controls an estimated 85% of the humanoid robot market. Unitree and AGIBOT each shipped more than 5,000 units in 2025, out of roughly 15,000 global shipments.
Why it matters:
US officials say foreign-connected robots could collect sensitive data, surveil facilities, or be remotely commandeered during a cyberattack. The policy may strengthen domestic robotics manufacturing, but it could also limit access to affordable hardware used by American startups, universities, and researchers.
How 2M+ Professionals Stay Ahead on AI
AI is moving fast and most people are falling behind.
The Rundown AI keeps you ahead of the curve.
It's a free AI newsletter that keeps you up-to-date on the latest AI news, and teaches you how to apply it in just 5 minutes a day.
Plus, complete the quiz after signing up and they’ll recommend the best AI tools, guides, and courses — tailored to your needs.
OPENAI
Image Source: Reuters
OpenAI says an internal version of Astra, its next major model, generated new results for 10 long-standing problems across mathematics and theoretical computer science using roughly $2,000 worth of tokens at Sol API rates.
The Details:
The results span sphere packing, coding theory, group theory, operator algebras, arithmetic circuit complexity, quantum computing, lattice cryptography, and extremal combinatorics.
Astra constructed the first known non-sofic groups and disproved Connes’s rigidity conjecture, addressing major open questions in group theory and operator algebras.
It also produced improved bounds for high-dimensional sphere packing and binary codes, plus new hardness results for the closest vector problem, which is relevant to post-quantum cryptography.
Three results resolve Erdős problems involving multicolor Ramsey numbers and extremal graph theory, while another settles Ehrhart’s volume conjecture across every dimension.
Astra generated the mathematical arguments, humans used the model to prepare them as research manuscripts, and the model then formalized each result as a machine-checkable Lean certificate.
Why it matters:
This goes beyond solving benchmark questions because Astra produced original arguments on problems that had remained open for years. The results still need scrutiny from the wider mathematics community, but they point toward AI becoming a practical research collaborator in advanced science.
ALIBABA
Image Source: Alibaba
Alibaba has launched Qwen3.8-Max, its largest and most capable flagship model yet, featuring 2.4 trillion total parameters, native multimodal intelligence, and a context window of up to 1 million tokens.
The Details:
Qwen3.8-Max uses a sparse Mixture-of-Experts architecture, activating only 95 billion parameters per request to improve efficiency and reduce inference costs.
The model combines sparse MoE with a hybrid attention system, helping it process extremely long inputs while maintaining performance across complex tasks.
Alibaba says Qwen3.8-Max performs strongly across coding, research, multimodal reasoning, and autonomous agent workflows.
The model reportedly completed a 16-day software engineering project independently and can turn 200-page documents, full TV series, or 100-hour livestreams into structured, searchable knowledge.
Qwen3.8-Max is now available globally through Alibaba Cloud Model Studio and QwenWork, with open weights expected to arrive on Hugging Face soon.
Why it matters:
Qwen3.8-Max combines trillion-scale capacity with sparse activation, making it more efficient than a similarly sized dense model. Its upcoming open-weight release could give developers access to one of Alibaba’s most advanced coding, multimodal, and long-horizon agent models.
AI TUTORIAL
Stop doing repetitive work manually — Uplift builds, runs, and maintains custom AI agents for your team so you never have to.
Steps to Follow:
Go to getuplift.ai and request early access
Open the scoping chat and describe your routine in plain language (takes 5–15 mins)
Review the workflow map Uplift generates — showing every trigger, integration, and decision branch
Approve it, and Uplift's team builds the full agent for you (no coding, no debugging)
Your agent goes live, monitored 24/7 with every run logged automatically
Sit back — Uplift handles all updates, API changes, and maintenance forever
Most AI tools make you think. Uplift makes AI actually do the work — so your team gets their hours back.
MORE AI & TECH NEWS
🗂️ Perplexity Renames Spaces to Projects
Perplexity has renamed Spaces to Projects, bringing chats, files, Computer tasks, custom instructions, and collaboration into one workspace with shared context and searchable project history.
🏗️ $15B Loan Talks Back Anthropic AI Campus
Morgan Stanley-led banks are reportedly discussing a $15 billion loan for a Texas AI campus serving Anthropic, with Google expected to back the project through lease guarantees and a reported 20% stake.
🛡️ White House Eyes Voluntary AI Safety Testing
OpenAI CEO Sam Altman is expected to discuss voluntary cybersecurity testing for advanced AI models with White House officials as the U.S. develops a new AI safety framework.
👔 Francis deSouza Becomes Scale AI CEO
Scale AI has named former Google Cloud executive Francis deSouza as CEO, leading the company's expansion from AI training data into enterprise AI deployment software.
🌍 Google Pauses AI Image Generation in Earth
Google has temporarily disabled its new Earth-based AI image generation feature after users created misleading scenes using real-world locations, prompting policy concerns.
🇨🇳 Chinese Military Researchers Use Frontier AI Outputs
Reuters found Chinese military-linked researchers using outputs from OpenAI and Anthropic models to help train smaller local AI systems for defense and cyber-related applications.
🎥 MiniMax Launches Open-Weight 2K Video Model
MiniMax introduced H3, an open-weight video model supporting 2K generation with stereo audio and multimodal inputs, while promising lower costs and broader hardware compatibility.
⚖️ German Court Rules Against Suno in Copyright Case
A German court ruled that AI music company Suno infringed copyrighted works represented by GEMA, ordering revenue disclosures and adding pressure for licensed AI training data.
TRENDING AI TOOLS
VidLux: Turn any image into a cinematic AI video.
UGCVideo: Generate realistic UGC ads without hiring creators.
Seedeo: Edit videos by simply describing the changes you want.
Freebeat: Turn any song into viral dance and music videos.
RespondThat: Let AI handle emails and phone calls for you.
Thrum: Create personalized songs for any person or occasion.
ReqBrief: Convert client conversations into polished project briefs.
AI MEME OF THE DAY
SPONSOR US
AI Insights newsletter is read by thousands of AI and Tech professionals/enthusiasts around the world.
Get in touch to get your product seen today!
Or email us at: [email protected]
THANK YOU FOR READING
FEEDBACK
How would you rate today's newsletter?
We'd love to hear your feedback or any interesting thoughts you have!
Please share by replying to this email.
