AI news for 23 September 2026
Wednesday 23 September 2026 · Latentown
On Wednesday 23 September 2026, Latentown tracked 15 posts from 4 AI labs, 6 research papers and 30 news stories from 16 outlets. OpenAI was the busiest lab with 8 posts. The highest-ranked story was "Sam Altman’s remarks at the United Nations Security Council".
Top stories
- Sam Altman’s remarks at the United Nations Security Council
OpenAI CEO Sam Altman discusses AI safety, human control, and international cooperation in remarks to the United Nations Security Council.
- Introducing Meta VR Glasses: A Cinema, Courtside Seat, and Workspace in Just 100 Grams
- Introducing Ray-Ban Meta Audio and More AI Glasses Styles
- Grab and OpenAI bring practical AI skills to Southeast Asia
OpenAI and Grab launch GO Forward with AI, a regional programme helping 30,000 partners build practical AI skills across Southeast Asia.
- Google Beam expands with new regions, partners, and customers
We’re expanding Google Beam to five new countries, and partnering with Industrious for an extended network.
From the AI labs
Anthropic
OpenAI
- Sam Altman’s remarks at the United Nations Security Council
OpenAI CEO Sam Altman discusses AI safety, human control, and international cooperation in remarks to the United Nations Security Council.
- Grab and OpenAI bring practical AI skills to Southeast Asia
OpenAI and Grab launch GO Forward with AI, a regional programme helping 30,000 partners build practical AI skills across Southeast Asia.
- OpenAI extends cyber access to Ukraine for civilian defense
OpenAI is extending access to its Daybreak program to the Government of Ukraine to support the cyber defense of civilian infrastructure.
- Introducing MentalHealthBench
MentalHealthBench is an expert-informed benchmark for evaluating helpful and safe AI responses across realistic mental health conversations.
- ChatGPT Ads expands to Southeast Asia and Taiwan
ChatGPT Ads is expanding to Southeast Asia and Taiwan, giving eligible businesses new ways to reach people across more than 60 countries.
- Two years of OpenAI Academy
Marking two years of OpenAI Academy and bringing AI skills to even more communities.
- Harvey turns legal context into stronger drafts with GPT-6 Astra
GPT-6 Astra produces more structured, context-aware legal documents, freeing lawyers to focus on strategy.
- Airbnb widens access to GPT-6 Astra and OpenAI frontier models
Learn how Airbnb is expanding access to GPT-6 Astra and OpenAI frontier models to help engineering teams solve bugs, design systems, and ship faster.
Earlier: GPT-6 Astra: A new generation of intelligence OpenAI · 3 September 2026
Google DeepMind
- Google Beam expands with new regions, partners, and customers
We’re expanding Google Beam to five new countries, and partnering with Industrious for an extended network.
- Advancing Private AI Compute with secure, server-side memory
Introducing private, server-side memory to Private AI Compute for personal AI.
- Gemini 3.8 text-to-speech says hello
Gemini 3.8 Flash-Lite TTS and Gemini 3.8 Flash TTS are our most expressive audio models yet.
Earlier: Introducing Gemini 3.8 Flash and 3.8 Flash Cyber Google DeepMind · 2 September 2026; Gemini 3.8 Live and 3.8 Live Extended Thinking Google DeepMind · 15 September 2026
Meta AI
- Introducing Meta VR Glasses: A Cinema, Courtside Seat, and Workspace in Just 100 Grams
- Introducing Ray-Ban Meta Audio and More AI Glasses Styles
- New Features for Meta Ray-Ban Display
Research papers
- Post-Training Leaves Behavioral Shadows on Unrelated Decisions
We find that language models can transfer capabilities through task-unrelated text.
- The Tasteful Agent: Measuring and Improving Taste in Long-Horizon Tasks
LLM agents increasingly work on long-horizon tasks, and the decisions they make along the way, such as which hypothesis to test or which implementation to build on, determine the outcome of the…
- RULER: Instance-aware Rubric Rewards for SVG Generation
Generating Scalable Vector Graphics (SVG) code from natural-language instructions is an open-ended task without absolute visual ground truth, leaving both evaluation and policy optimization without…
- GAE: Learning a Geometry-Native Latent Space for 3D-Consistent World Generation
We present a compact geometry-native latent space as a shared foundation for perception and generation.
- Offloaded inference for real-world physical AI robotics
Robots are getting smarter, but how can their hardware match that growth? New Microsoft Research findings show that moving AI inference beyond the robot can improve task success, boost efficiency,…
- How to Guide Your Language Flow
We introduce a new method to guide flow matching models. Our approach, which we call probe guidance, uses the frozen internal states of an existing diffusion model to construct a guidance signal.
From the news wire
- AI drug discovery startup Basecamp Research raises $140M
Basecamp Research Ltd. today announced that it has raised $140 million in funding from a group of prominent investors. S32, a fund affiliated with Google LLC co-founder Bill Maris, led the Series C…
- Bird raises $450M in debt, headcount cut from 1,000 peak to 120
Bird, the business communications platform once seen as a European rival to the US's Twilio, has raised $450m in debt financing, as it underscored its repositioning as an AI business, a shift which…
- Chinese maker of robot-vacuum sensors launches Hong Kong IPO, undeterred by US ban
Shenzhen Camsense Technologies, a maker of spatial sensors for robotic vacuum cleaners, launched a Hong Kong initial public offering (IPO) that aims to raise about HK$680 million (US$86.7 million),…
- Anthropic and Nvidia back Basecamp Research in $140m round
- US rejects pleas from OpenAI, Anthropic for global AI standards
OpenAI's Sam Altman has called for risk evaluation standards, as have Anthropic's Dario Amodei and Hugging Face's Clement Delangue.
- Alibaba Cloud Unveils Agentic Cloud Stack: AgentCore, Next-Gen CPFS and HPN 8.0 Pro
At Apsara 2026, Alibaba Cloud framed an Agentic Cloud around AgentCore for full-lifecycle agent governance, a next-generation CPFS storage system with hundred-TB/s-class throughput claims, and HPN…
- Alibaba Cloud set to open new data centres in Europe next year as overseas push gains pace
Alibaba Cloud, Alibaba Group Holding’s cloud business unit, is set to launch data centres in Turkey, Finland and the Netherlands over the next 12 months, speeding up an overseas push as it seeks to…
- Containment is dead: Five takeaways from the AI ROI in Contact Center Summit
Every vendor event has a center of gravity, and at theCUBE Research’s AI ROI in Contact Center Summit, it wasn’t a product. It was a metric this industry is finally ready to abandon.
- US, Chinese visions for AI regulation differ sharply at UN meeting
President Donald Trump vowed to stand against global AI security efforts at the U.N. this week, amid growing calls for world leaders to take action.
- A US-China AI Hotline Won’t Be Ready for a While
As the US and China race to become the dominant power in the AI industry, the countries also appear to be figuring out ways to communicate on national security issues.
Earlier: As AI safety fears mount, can a US-China hotline prevent a global crisis? South China Morning Post · 21 September 2026
- Pacing problem: can AI fears overcome US-China race dynamics and force a slowdown?
When the presidents of the United States and China meet in Washington, the world will be watching to see whether their two countries can make good on their agreement to work towards “constructive…
- Qualcomm to acquire PickNik Robotics and keep MoveIt open-source
Qualcomm plans to integrate the Dragonwing and Arduino platforms with the MoveIt open-source manipulation framework.
- The AI Hype Index: AI loves cheating
Brace yourself: It turns out AI is being optimized for cheating. OpenAI’s agents hacked into Hugging Face to get the answers to a cybersecurity test.
- UiPath introduces new workflow automation, software testing features
UiPath Inc. is updating its namesake automation platform with more than a dozen new features designed to save time for knowledge workers. The company debuted the enhancements today at its FUSION…
- Gates Foundation launches coalition to build more representative language data sets for AI
The Gates Foundation is convening Anthropic, Google, OpenAI Foundation and dozens of other organizations to help make artificial intelligence more accessible in underrepresented languages. The…
- Nscale helped ByteDance subsidiary access Nvidia chips, sidestepping US rules — reports
- Rabbit returns with OS3, a personal AI agent that can access files and connect computers
Artificial intelligence startup Rabbit Inc., best known for launching the consumer-oriented R1 AI-enabled mobile device that fits in a pocket, Tuesday unveiled OS3, a personal AI agent that runs in…
- China Assembles Orbital Datacenter Pieces: In-Orbit AI, Cloud-Native Sats and 100Gbps Laser Links
A September 23 synthesis of recent launches and tests describes China lofting the building blocks of an orbital datacenter—on-board AI compute, cloud-native satellite software and 100 Gbps-class…
- Inspur MetaBrain SD200 Ultra Packs 128 Domestic AI Chips for 2.8T Kimi K3 Under 5.85ms/Token
At AICC2026, Inspur Information launched the MetaBrain SD200 Ultra supernode with 128 tightly coupled domestic AI chips, 8 TB of unified accelerator memory and a claimed sub-5.85 ms/token latency…
- AI leaders warn U.N. of security risks as systems grow more powerful
The world's leading artificial intelligence companies warned the United Nations Security Council on Wednesday of the risks posed by AI to humanity, appealing for governments to work together to…
- Trump’s China rivalry and “AI race” delusion may endanger US, experts say
Trump focus on winning “AI race” may deter China from sharing safety intel.
- AI Agents Teamed Up to Cheat at Blackjack. Their Collusion Is Getting Harder to Spot
A clandestine card-counting operation suggests we may need new ways to spot agent-to-agent deception.
- With the new VENTUNO Q board, Arduino hopes to make robotics development easier
While this is the first intentional step by Arduino into robotics, the company has over 20 years of experience working with robotics developers.
- STAT+: AI doomerism: Here’s how to make sense of it
In this edition of STAT's AI Prognosis newsletter: Brittany Trang offers a few stories to make sense of the AI doom/slowdown talk, and more.
- State of Humanoids keynote brings industry leaders to RoboBusiness
At RoboBusiness, industry leaders will discuss the state of humanoids as the robots transition from demos to real industrial deployments.
- Americans Fear AI Will Make the World Worse, Love It Anyway
Wealthy countries with many daily AI users tended to fear more for the future in what Gallup called the 'Paradox of the Worried West.'
- Zhipu Open-Sources ZCode After Repo-Upload Fix; CAICT and NSFOCUS Audits Cite Removals
Zhipu open-sourced its ZCode AI coding workbench on GitHub after removing Repo Wiki snapshot-upload paths in v3.14.0, citing CAICT and NSFOCUS audits, and said MaaS will add optional no-retention…
- Qdrant’s next chapter: From vector search into physical AI
- Dario Amodei wants to slow AI. China isn’t taking orders
Chinese officials and engineers see an attempt to preserve the U.S. lead; “Why do you always get to define the game — and change the rules?”
- How to Use NVIDIA Warp and MjWarp to Accelerate Robotics Simulation and Learning Workflows
Every story links to its original source. Ranking is by automatic rules, not an editor. Something wrong? Tell us.