Welcome to the DailyAI Briefing, the day’s five biggest AI stories.
Let’s dig in.
$79.38 billion from the company’s own share sale
$79.38 billion: that is the gross proceeds, before offering fees are deducted, from the 956,351,354 Class A shares the company itself is selling. The offering as a whole is 980 million shares, of which 23,648,646 are being sold by existing stockholders. The company will not receive the proceeds from those sales. The official announcement prices the offering at $83 a share. That is above the earlier $70 to $78 range.
The ticker is ANT, on the Nasdaq Global Select Market. The $83 figure is the offering price, not a stock-market trading price. The offering is expected to close on 20 November, subject to customary conditions, and the underwriters have a 30-day option to buy up to another 147 million shares. A registration statement was declared effective on 18 November. Our report on the pricing sets out the share counts and the trading date.
We’ve priced our IPO at $83 per share, with trading expected to begin on Nasdaq tomorrow. https://t.co/igHvzweKdb
— Anthropic November 18, 2026
A 48-hour deadline for the most serious incidents
48 hours: that is the reporting deadline California’s Government Operations Agency now wants for most critical AI safety incidents, down from 15 days. The same report, dated 16 November and released by the governor’s office on Wednesday, also asks lawmakers to let the state order an immediate shutdown of a model if officials believe it poses an imminent catastrophic or major loss-of-control risk and the developer refuses to act. The governor’s announcement presents the document as a plan for new legislation, not a law that is already in force.
As proposed, a shutdown order would take effect at once. The firm that built the model could ask a court to pause it, but the order would stay in place until that review finished. Every large firm that builds the most capable models would also have to keep a tested plan to shut a model down at several levels. The report says the state, not the companies, should set baseline safety standards, and that licensed independent inspectors would share unredacted assessments with the state. These remain recommendations until lawmakers act. Our report on the recommendations covers the proposed shutdown order and the checks on those firms.
A new model at $3 per million input tokens
$3 per million input tokens, and $15 per million output tokens: that is the list price for Gemini 4 Neon, a new model introduced on 18 November for complex software engineering, knowledge work and science workflows. Cached input tokens are priced at $0.15 per million, a 95% discount, and the context window is one million tokens. The company places Neon between Gemini 4 Carbon, which it says is for the hardest, long-horizon projects, and Gemini 4 Hydrogen, which it says handles fast everyday work. The official product post says the model is rolling out in the paid API, for business users, and to paying subscribers in the consumer app and a coding workspace.
The company said that, at the model’s highest thinking level, it ranks third overall with a score of 61 on the Artificial Analysis Intelligence Index, and second on the Vals Index with 70.9%. It also claimed first place on the Artificial Analysis Coding Agent Index, with a score of 58.6. Those figures come from the company’s announcement and have not been independently verified here. Our report on the launch records the prices, the rollout and the company’s own scores.
Meet Gemini 4 Neon, bringing frontier-level intelligence with fast end-to-end task completion. Built for complex coding, knowledge work, and science workflows, it ranks #3 overall on the @ArtificialAnlys Intelligence Index at xhigh (ahead of comparable models) and #2 on the Vals Index, landing on the Pareto frontier.
— Google November 18, 2026
A 2029 target for rack-scale satellites in orbit
The first quarter of 2029 is the target for the company’s first dedicated heavy-lift launch under Project Suncatcher, a research plan to put solar-powered satellites with roughly the computing capacity of a data-center rack into a sun-synchronous orbit. The official research post, dated 19 November, sets out that launch window and the rack-scale payload. The flight would use New Glenn, a heavy-lift rocket, so the satellites can stay in sunlight and run on solar power collected in space.
The satellites would carry Tensor Processing Units, custom chips the company designs to run AI models. The longer-term aim is an interconnected network of such satellites, not a single machine. A prototype reached orbit on 1 October on a shared flight, and on 30 October the team said it received the first response from a data-center-class chip in orbit. A 2029 flight would test whether some of this work can be done in space at all. It would not replace the large data centers now being built on the ground. Our report on the launch plan covers the orbit, the chips and the prototype already in space.
Blue Origin and Google have entered into a launch services agreement for Project Suncatcher, Google’s research moonshot exploring how an interconnected network of solar-powered satellites equipped with its AI chips, Tensor Processing Units (TPUs), could scale machine learning compute in space. https://t.co/84OZUsKBsI
— Blue Origin November 19, 2026
A £400 billion gain that hangs on ten safety gaps
£400 billion: that is how much larger a House of Commons committee says the UK economy could be after ten years, if ministers close ten gaps in how the most powerful AI systems are tested and stopped. The figure is the committee’s conversion of an OECD estimate of 0.4 to 1.3 percentage points a year in extra UK labour productivity growth, or up to £38 billion of additional output a year. The report, published on 18 November, is a set of recommendations, not a new law. Ministers have two months to reply.
The MPs say the UK has no law governing the safety of frontier AI, meaning the most capable general-purpose models, and that the government’s testing body is not a regulator and relies on voluntary access. Among the ten gaps are no legal right to inspect models, no independent check of safety claims, no duty to report incidents, and no power for a public body to halt a deployment. The committee wants a new safety and security bill in the next parliamentary session, and a statutory regulator kept separate from the testing body. Until ministers accept or reject the recommendations, the testing body will still rely on companies choosing to cooperate. Our report on the inquiry sets out the ten gaps and the proposed new law.
That’s all for today. See you tomorrow.
The post DailyAI Briefing: Anthropic IPO at $83 a share, plus 4 more appeared first on DailyAI.