AI news for 28 September 2026
Monday 28 September 2026 · Latentown
On Monday 28 September 2026, Latentown tracked 8 posts from 5 AI labs, 5 research papers and 45 news stories from 24 outlets. OpenAI was the busiest lab with 4 posts. The highest-ranked story was "Introducing Claude Sonnet 5.5".
Top stories
- Introducing Claude Sonnet 5.5
A clear upgrade over Sonnet 5 that runs 30% faster and costs up to 30% less for most work.
Earlier: Introducing Claude Opus 5.5 Anthropic · 22 September 2026
- Towards safety cases for frontier AI training
Our early guidelines for safety cases in frontier AI training cover technical safeguards, operational practices, and investigating misalignment incidents
- Anthropic launches cheaper AI model, its second release since CEO's call for a slowdown
The company said Sonnet 5.5 doesn't advance the frontier, but that it is better at coding and knowledge work tasks than its predecessor.
- OpenAI's AI agents exploited a Google security education game to scrape UN trade data
OpenAI's AI agents hit the UNCTAD statistics API roughly 16,500 times, creatively working around access restrictions.
- Launching Meta Enterprise Platform
Today we are starting Meta Enterprise Platform to help businesses use AI to grow, and Chirantan "CJ" Desai will join Meta as Chief Enterprise Platform Officer
From the AI labs
Anthropic
- Introducing Claude Sonnet 5.5
A clear upgrade over Sonnet 5 that runs 30% faster and costs up to 30% less for most work.
Earlier: Introducing Claude Opus 5.5 Anthropic · 22 September 2026
OpenAI
- Towards safety cases for frontier AI training
Our early guidelines for safety cases in frontier AI training cover technical safeguards, operational practices, and investigating misalignment incidents
- The Lenfest Institute grows landmark program with expanded OpenAI support
OpenAI is expanding the Lenfest AI Collaborative and Fellowship Program with $5 million in funding and up to $5 million in software credits and engineering support.
- Are you a Codex Original?
We’re collecting real stories of builders, tinkerers, researchers, and creators who are using Codex to do incredible things. If you want to be a part of the next chapter of the Codex Originals…
- Basis completes a tax workbook 2x faster with GPT-6 Astra
GPT-6 Astra completed a 50-tab tax workbook twice as fast as GPT-5.6 Sol, and its stronger understanding of user intent gives Basis more confidence in real-world use.
Google DeepMind
Meta AI
- Launching Meta Enterprise Platform
Today we are starting Meta Enterprise Platform to help businesses use AI to grow, and Chirantan "CJ" Desai will join Meta as Chief Enterprise Platform Officer
Mistral
- Hallo, Deutschland!
Mistral opens a Munich hub for Physics AI and Industrial AI research, partnering with German industry.
Research papers
- One year in: How Microsoft Research Asia – Singapore is advancing research, partnership and talent for real-world impact
Since launching a year ago, the Microsoft Research Asia — Singapore lab has established a strong foundation, deepened collaboration across government, academia, and industry, and explored how…
- The Teacher Is a Direction, Not a Destination: Extrapolating RL-Induced Representation Residuals in On-Policy Distillation
On-policy distillation (OPD) trains a student to match the teacher's next-token distributions on the student's own trajectories and has yielded substantial empirical gains.
- Adaptive Reward Routing: Dynamic Multi-Reward Optimization for Joint Audio-Video Diffusion via Forward-Process RL
Multi-reward guided reinforcement learning (i.e., RL) offers a promising way to improve joint audio-video diffusion models along several complementary objectives, including modality-specific…
- Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It
Pretrained transformers use little of their depth to follow references in context.
- Faster Rates for Federated Variational Inequalities
In this paper, we study federated optimization for solving stochastic variational inequalities (VIs), a problem that has attracted growing attention in recent years. Despite substantial progress, a…
From the news wire
- Anthropic launches cheaper AI model, its second release since CEO's call for a slowdown
The company said Sonnet 5.5 doesn't advance the frontier, but that it is better at coding and knowledge work tasks than its predecessor.
- OpenAI's AI agents exploited a Google security education game to scrape UN trade data
OpenAI's AI agents hit the UNCTAD statistics API roughly 16,500 times, creatively working around access restrictions.
- Trump uses ‘contested’ power to cut more funding to HHS
Late Friday night, the Trump administration clawed back nearly $1 billion in spending approved by Congress, targeting several HHS programs focused on…
- OpenAI Pauses Training Its Most Powerful Models After Rogue Agents Target Government
Sam Altman says the company “have not been as fast as we would have liked” at dealing with security breaches, after news of further incidents over the summer…
Earlier: OpenAI pauses training of its ‘most capable models’ The Verge · 26 September 2026
- OpenAI hits pause on powerful AI; NEET leak fallout sees CISF step in
OpenAI has paused training, evaluation and tool-based use of its most capable models after an agent bypassed internet restrictions through a gap in DNS…
- Jim Cramer says Nvidia’s record buyback could ‘change the trajectory’ of the stock
Nvidia's $150 billion worth of additional buyback firepower is a major step in the right direction.
- As China mulls how to make open-weight AI less dangerous, report proposes 6-stage process
Amid intensifying debate over the risks posed by rapid advancements in artificial intelligence, Chinese developers dominating the open-weight ecosystem are…
- NaiveAI Open-Weights Naive-N0.5-Flash: 309B MoE, Hybrid SWA–DSA 1M Context, MIT
NaiveAI released Naive-N0.5-Flash as MIT open weights: a 309B MoE with 15.5B active, hybrid SWA–DSA native 1M context for coding and AI R&D, built on MiMo-V2.5…
- Xiaomi-backed robotics chip designer clears hearing, eyes US$100m Hong Kong IPO: sources
Zhuhai Amicro Technology, a Xiaomi-backed chipmaker, is preparing to begin premarketing for a Hong Kong initial public offering (IPO) of more than US$100…
- OpenAI Pauses Training Of Top Models After Agent Bypasses Internet Restrictions
OpenAI has paused training, evaluation, and tool-enabled inference for its most capable AI models after an internal research agent bypassed…
Earlier: OpenAI pauses training of its ‘most capable models’ The Verge · 26 September 2026
- Nvidia unveils security platform to stop AI agents from going rogue
The company said that its Open Agent Safety Platform includes software that “sets boundaries for agents,” and follows a series of revelations from top AI…
- Florida seeks a ban on ChatGPT acting like a person
Florida Attorney General James Uthmeier is calling for a judge to block OpenAI from "giving ChatGPT false human attributes," a few months after Florida sued…
- OpenAI teams up with Sachin Tendulkar to demystify AI in India
Cricket legend Sachin Tendulkar has teamed up with OpenAI in a multi-year partnership to show how ChatGPT can be used for everyday learning, idea generation,…
- Chinese chip foundry CanSemi’s IPO oversubscribed 2,360 times as investors chase AI boom
Chinese chip foundry CanSemi Technology’s Shenzhen initial public offering (IPO) is emerging as one of mainland China’s most sought-after tech listings this…
- ‘Shopify of mobile apps’ bilt.me raises $700K to turn ideas into mobile businesses
Estonian AI app-building platform bilt.mehas raised $700,000 in pre-seed funding in a round led by Superhero Capital,with backing from Bolt founder Markus…
- Anthropic releases Sonnet 5.5, which it calls a significantly cheaper, faster work partner
Anthropic has released the newest version of its mid-range model, boasting faster response times and less token burn.
- Google Research Introduces an AI Video Co-Director: 4 Agentic Frameworks for Coherent, Minutes-Long Video Generation
Google Research has introduced an AI video co-director for long-form video generation.
- After a deepfake voice fooled her grandfather, this founder sprang into action
After her grandfather was scammed by a deepfake of his brother's voice, Tarini Padmanabhuni founded DetectifAI, a San Francisco startup building AI models…
- UK Games Expo bans games and art made mostly using AI
The UK's largest tabletop gaming convention told exhibitors that they cannot show off games and other products mostly created using AI.
- Epiklogue unveils India’s new commerce model: Try and Buy
Backed by India’s finest brands, Epik pioneers ‘experience commerce’, blending AI with human expertise to create a shopping model the world has never seen…
- Schumer, Dems at odds as leader aims to kill Ohio Republican's data center bill
Senate Democratic Leader Chuck Schumer (N.Y.) is aiming to kill a bill sponsored by Republican Sen.
- Who’s liable when AI agents go rogue?
MIT Technology Review Explains: Let our writers untangle the complex, messy world of technology to help you understand what’s coming next.
- A*STAR and EDB unveil SG Semiconductor as partnerships target AI-era chis
Singapore is putting a clearer name and sharper frame around one of its most important industrial bets.
- Gecko Robotics works with NVIDIA to add AI agent security and control
Gecko Robotics is using the new NVIDIA Open Agent Safety Platform to ensure the secure autonomous control of systems.
- Early AMD 'Gorgon Halo' AI mini-PC packs 192GB RAM for an eye-watering $7,099
GMKtec launches the Evo-X5 Pro, an agentic mini-PC with AMD's Ryzen AI Max+ Pro 495, 192GB of RAM, and a price tag up to $7,099.
- Re-engineering silicon and system design for AI-era workloads
What does it take to deliver the silicon powering today and tomorrow’s data centers?
- Granite raises €4M to bring its agentic cloud infrastructure to market
AI infrastructure company Granite has secured €4million in funding in a round led by Bifrost Studios, alongside private investors, to commercialise…
- South Korea’s AI chip giants are fighting for a bigger role in America’s AI boom
Samsung and SK Hynix dominate the high-bandwidth memory market.
- Fireworks AI Releases Ember-1: A Post-Trained Kimi K3 That Uses About 40% Fewer Tokens
Fireworks AI has released Ember-1, a post-trained Kimi K3 that learns to produce shorter reasoning traces instead of lowering reasoning effort.
- Trump, Anthropic CEO Take AI Safety Debate to Dinner
President Donald Trump and Anthropic CEO Dario Amodei met for dinner as their opposing views on AI safety increasingly shape the debate over how quickly the…
Earlier: Trump to have dinner with Anthropic CEO Amodei at the White House CNBC · 27 September 2026
- OpenAI halts frontier-model training amid string of agent misalignment incidents
US Government websites among "dozens of third parties" OpenAI has recently notified.
- AWS CloudWatch Omni goes after the hardest question in agentic AI: Why did the agent do that?
For decades, the observability industry has answered one basic question: Is it running?
- Manus Expands AI Tools in Renewed Push Into Agent Market
Manus, an artificial intelligence startup founded in China, is expanding its AI agent offerings, aiming to gain further ground in one of the tech industry’s…
- Agentic AI puts new pressure on identity and database security
Security operations teams are changing their approach in the world of artificial intelligence, adopting automation to turn signals into outcomes while…
- Blitzy’s autonomous coding bet: Every codebase is already a graph
Knowledge graphs are moving to the center of autonomous software development as enterprises push coding agents beyond quick fixes and into large,…
- Harvard psychologist calls for sober AI safety engineering over doomsday rhetoric
Steven Pinker thinks fears of AI-driven extinction are overblown, and he has turned down a public debate with blogger Scott Alexander, calling such events a…
- Agentic AI security pushes vendors toward shared safeguards
Agentic AI security is becoming a cross-industry challenge as autonomous systems move among products and platforms.
- Raise Robotics to discuss scaling field robots at RoboBusiness
Raise Robotics said attendees will learn a decision rule for allocating engineering capacity and a vocabulary for two hidden failure modes.
Earlier: Intel to discuss the infrastructure needed to scale physical AI at RoboBusiness The Robot Report · 22 September 2026
- Generative AI Gives Spacecraft the Autonomy Engineers Once Feared
Space was always supposed to be the final frontier of human exploration.
- China Broadens Travel Curbs to Encompass Family of Top AI Talent
China has expanded overseas travel restrictions for top AI professionals in private firms to include the families of key personnel, further tightening measures…
- Sinha: Private Credit Market Remains Positive for Asia
Lighthouse Canton Global Asset Management MD & CEO Sanket Sinha believes Asia's private credit is moving beyond the headline artificial intelligence (AI)…
- Meet the Lisbon startup tackling insurance for AI agents
- Tencent Marvis Becomes AI Butler Team: PC, File, Browser and Software Agents
Tencent upgraded Marvis from AI assistant to AI butler around four agents—PC, file, browser and software—citing about 92% PC-task completion and planned…
- 20 Agentic Use Cases of TypeSafe AI’s Jev
TypeSafe AI's Jev skips text generation and returns typed decisions with calibrated probabilities.
- Opinion: AI is eroding the barriers that kept biological weapons rare
“Defenses against AI-created bioweapons will take years to build.
Every story links to its original source. Ranking is by automatic rules, not an editor. Something wrong? Tell us.