← Dailies

Daily · 2026-09-21 00:00 UTC

AI labs face antitrust suit over slowdown pact; Anthropic delays IPO — 2026-09-21 00:00

An antitrust lawsuit targets Anthropic, OpenAI, SpaceXAI and Google over an alleged agreement to slow AI development; Anthropic reportedly pushes its IPO to November; Runway demos real-time video generation; and Alibaba open-weights a 7B image model.

Source ↗

News Digest — 2026-09-21 00:00 UTC — 10 items

1. Anthropic, OpenAI, SpaceXAI and Google face an antitrust lawsuit over an AI slowdown agreement — Plaintiffs allege the four companies made an illegal agreement to slow AI development, saying the plan had been in motion for months and calling the agreement a "shortcut" that "substitutes collective restraint for individual accountability." Tom's Hardware · Sep 20
2. Anthropic reportedly pushes its IPO from October to November — The company originally targeted October; according to The Information and The Wall Street Journal the listing is now expected in late October at the earliest but more likely November, with advisors saying Anthropic wants to present strong third-quarter results before going public. Investors expect a valuation of roughly $2 trillion and a capital raise of up to $100 billion. Revenue more than doubled between the first and second quarters of 2026, from about $4.7 billion to more than $11.5 billion. The Decoder · Sep 20
3. Runway demos real-time AI video generation built on its GWM-1 world model — Instead of entering a prompt and waiting for a finished clip, users would stream video as they describe it. Runway says users repeatedly report losing the most time generating and revising, and it wants to minimize time to the first frame; it uses GWM-1, its first "General World Model" introduced in December 2025, which builds on Gen-4.5, generates video frame by frame, and accepts camera movements, robot commands or audio as controls. Runway points to interactive applications — education, gaming, robotics — as the biggest long-term use case. The Decoder · Sep 20
4. Alibaba open-weights Qwen-Image 2.1, a 7B image generation and editing model — The model unifies text-to-image generation and editing with just 7 billion parameters in its visual generation component, built on 32 Single-Stream DiT layers. It generates regular or transparent RGBA images from text, edits transparent layers, supports up to 10 reference images at once, and takes local edits via circles, painted annotations or masks. It ships under a Qwen research licence that bars commercial use. Hugging Face · Sep 20
5. OpenAI forecasts about $278 billion in negative free cash flow through 2030 — OpenAI expects negative free cash flow of $278 billion between this year and 2030 as it invests to expand its access to computing power, according to a presentation viewed by the Financial Times. Revenue is forecast to rise from $36 billion this year to $350 billion in 2030, for $840 billion in total revenue across the period, and the company is in talks for a new round at a valuation above $1.2 trillion. PYMNTS · Sep 20
6. The world-model labs are keeping their roadmaps secret — TechCrunch's Russell Brandom moderated a panel on world models at the All In conference and found the field's biggest names unwilling to say what they are building: Yann LeCun's AMI Labs and Fei-Fei Li's World Labs. AMI Labs co-founder and VP of World Models Michael Rabbat said the company is "still in a research and building phase," telling the panel "We'll talk about it when we're ready to talk about it," while data supplier Physicl's chief executive Alex de Vigan said he does not know what his data is being used for. TechCrunch · Sep 20
7. Vocci puts AI meeting notetaking in a ring — Double-tap the button to start and stop recording; tap and hold to ask Vocci AI questions, which works only while the app is open. The ring weighs under 6 grams with titanium on its inner and outer surfaces, the company claims 8 hours of recording on a single charge, and it ships with a case that can recharge the ring up to three times. TechCrunch · Sep 20
8. Jensen Huang says there is a "0% chance" AI ends the world — In an interview with CBS Sunday Morning, Nvidia's chief executive claimed a zero percent chance of AI ending the world, said of people sounding the alarm that "Scaring people is unnecessary. It is irresponsible," and argued that calls from other tech leaders to slow AI development are "not grounded in science" and that no new rules, laws or guidelines are needed. The Verge · Sep 20
9. The energy grid's cyber risk is human and growing, not a rogue-AI story — Writing in The Verge, security researcher Joshua Corman of the Institute for Security and Technology argues energy systems were already disturbingly vulnerable to cyberattack well before recent high-profile hacks raised the prospect of AI agents orchestrating attacks, and that "any sociopath that wants to [attack] is now more powerful than they used to be." The Verge · Sep 20
10. Unit1 raises a near-£15m investment to recreate concerts with digital avatars — The UK music tech startup was set up last year by Barney Wragg, who ran Andrew Lloyd Webber's entertainment group for five years from late 2011. The Guardian reports a near-£15m investment and says the company secured $20m in funding from investors including Balderton Capital. It plans concerts fronted by "hyper realistic" avatars of living artists who no longer tour, and reconstructions of performances by dead musicians blending archive footage with filming actors and musicians who look or play like the departed star; it has tested the technology with KT Tunstall. The Guardian · Sep 20

Try This Next

• 🔧 Before you build on Qwen-Image 2.1, read the licence — a research-only term kills commercial plans, and the model runs on consumer-class GPUs
• 🎥 Watch Runway's real-time generation research and ask what you would build if video appeared as fast as you spoke
• 📝 Write down your own line for when an AI system should be allowed to keep its roadmap, or its incidents, secret — the world-model labs and this week's containment stories are the test cases

Content on Anagnorisis is summarized, paraphrased, and editorialized from publicly available sources for length and clarity. Original sources are linked where available. All trademarks belong to their respective owners.