Tuesday, 29 September 2026
The last 24 hours' hottest news in AI, sorted by field, each story with a line on why it matters.
Anthropic's IPO filing shows huge growth and huge losses
Anthropic's prospectus for its planned stock-market debut shows revenue jumping twelvefold to about $4.6 billion in 2025, and $11.5 billion in the second quarter of 2026 alone. It also shows an operating loss of more than $8 billion last year and plans to spend $518 billion on cloud and computing.
- 2025 revenue (12x growth)
- $4.6B
- Revenue in Q2 2026
- $11.5B
- 2025 operating loss
- $8B+
- Planned compute spending
- $518B
Why it's the story of the dayThis is the first detailed look at the finances of a top AI lab, ahead of what could be the largest IPO ever.
TLDR
Robotics
Robots, self-driving and AI in the physical world.
China made 77.9% of humanoid robot shipments in H1
Research firm IDC says China made 77.9% of the world's humanoid robot shipments in the first half of 2026. Chinese regulators are also setting tougher rules for robot companies that want to list on the stock market.
Why it mattersChina is well ahead of the rest of the world in actually shipping humanoid robots.
TechNode ↗U.S. Army picks a humanoid robot from 103 proposals
The U.S. Army chose one humanoid robot and four supporting technologies out of 103 proposals. It is one of the military's first formal steps toward testing humanoids.
Why it mattersMilitary interest could bring big money into humanoid robots.
Defence Industry Europe ↗Science & health
AI in biology, medicine, climate and maths.
Tremont AI launches to test drug safety earlier
Tremont AI came out of stealth with AI that reads several kinds of data, such as images and lab results, for preclinical drug safety testing. The aim is to spot unsafe drug candidates before they reach human trials.
Why it mattersCatching toxic drugs earlier saves years of work and lowers risk for trial volunteers.
Business Wire ↗Roche puts AI at the centre of its strategy
Roche laid out a plan to go from AI-assisted research to largely self-running labs. It is already using automated laboratories for drug discovery.
Why it mattersOne of the world's biggest drugmakers is betting that AI will change how medicines are found.
SWI swissinfo.ch ↗Open source
Open-weight models and community releases.
H Company releases Holo4 open-weight computer-use models
H Company released Holo4, a family of open-weight models that can click through apps, write code and call tools on desktop, web, Android and APIs. Anyone can download and run them.
Why it mattersFree models that can operate a computer let smaller companies build agents without paying a big lab.
MarkTechPost ↗
Google open-sources RRSI for self-improving agents
Google Research released RRSI, a framework that lets AI agents rewrite their own prompts and tools while the underlying model stays the same. It has checks so agents don't just get good at the test tasks and nothing else.
Why it mattersIt gives developers a free, tested way to let agents improve themselves over time.
MarkTechPost ↗Quick hits
Smaller stories worth a glance.
- OpenAI reopens $200 Pro plan but halves API creditsOpenAI reopened its $200-a-month Pro plan to new users but cut its included API credits in half.The Decoder ↗
- Alibaba's Qwen launches a real-time voice modelQwen-Audio-3.1-Realtime can listen and talk at the same time and decides for itself when to speak.MarkTechPost ↗
- Steven Pinker calls for practical AI safety workHarvard psychologist Steven Pinker says AI safety should focus on practical engineering and oversight rather than doomsday talk.The Decoder ↗
- Founder builds phone app to detect deepfake voicesAfter a fake voice fooled her grandfather, DetectifAI's founder built models that run on a phone and flag synthetic voices during a call.TechCrunch ↗
- Manus 2.0 adds video editing and phone-run agentsManus turned its AI agent into a platform, with a Studio for video editing and game building plus a Cue app that gives agents their own email address, wallet and computer.The Decoder ↗
On the calendar
What's happening in the next two days.
- 30 SeptConfirmed
Micron reports quarterly earnings
The memory maker's results are a check on how strong AI hardware spending still is.
- 1 OctConfirmed
So Connecticut AI and privacy laws take effect
Connecticut's new AI and data-privacy rules start on 1 October.
Stories are summarized in plain language; follow the link before quoting a figure. Story images are shown from the publishers' own sites.