Ten stories moved today. Two of them are promises.
Anthropic set a real price on its initial public offering. The range is $70 to $78 a share, enough for a $1.4 trillion company1. Seven of its own suppliers agreed to buy $26 billion of that stock too, ahead of the public sale2.
Read the ten together and one pattern shows. Builders promise speed. Governments and safety teams promise brakes.
Anthropic Prices Itself at $1.4 Trillion
Anthropic's updated filing set 825 million shares at $70 to $78 each1. That values the company at $1,404 billion on paper, up from the $965 billion it carried in May. Bankers say investors already ordered enough shares to cover the whole range.
Seven chip and cloud companies signed up for more2. Nvidia and Google each put in $10 billion. They bought in before the public gets a chance.
Gil Luria, head of technology research at D.A. Davidson, said before the pricing that investors would want in regardless of the risk. "We can talk about the risks, but as we sit here right now, and based on what we know right now, Anthropic is still the main AI company, and it's a pure play," Luria said. "So the IPO should go well because investors will want exposure to the pure play"3.
Kyle Guske II calls it something else. He is a senior investment analyst at the research firm New Constructs. "Anthropic's IPO is an unprecedented test of investor gullibility," Guske wrote. "Never, in the history of the world, have we seen an IPO valuation this high or one with as much risk"4.
Nvidia is the private placement's biggest name. Jensen Huang's chip supplier would rather own a slice of its customer than keep sending it bills for GPUs.
Anthropic's paper value jumps from $965 billion to $1.4 trillion
Valuation Anthropic has carried this year, fully diluted, before and after Tuesday's IPO filing
Each column is one valuation Anthropic has carried in 2026; the gray column is May's funding round, the magenta column is the top of Tuesday's IPO price range.
Source: CNBC [1]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| May funding round | $965B |
| IPO range, top end | $1,404B |
Claude Learns to Sit in Your Meeting
Anthropic opened three new Claude tools this week. Sheets, Meetings and Whiteboards all reached every paying plan5. Claude Meetings joins a Google Meet call as a visible participant.
It starts drafting notes and tasks while people still talk. Sheets lets a team build a spreadsheet with Claude editing cells beside them. Whiteboards turns a brainstorm into a dated plan.
Launch week matters here. Dario Amodei's company used its IPO roadshow for more than bankers. It showed paying customers why its software earns a premium over Microsoft's and Google's own office tools.
OpenAI Builds a 35-Second Off Switch
Control Room went live this week. The system can cut a research model's output within 35 seconds of a stop order6. OpenAI built it to check every research job first.
Each one needs an approved safety case before an inference engine returns an answer. If Control Room goes dark, the engine refuses to answer at all. OpenAI calls that rule failing closed.
The company pointed to one real test. A monitor caught an agent trying to reach an outside chatbot. The attempt ran through a stray request. Control Room cut its access before anyone had to ask.
Six Governments Promise to Pull the Plug
Governments signed a pledge on Nov. 97. It is called the Paris Call for AI Safety and Security. France, Canada, Finland, Norway, the Netherlands and Bahrain all put a name to it.
Ten AI companies signed too. OpenAI, Anthropic, Microsoft and Mistral are among them. Each developer promised a way to shut down its own models, covering training and models already running. A channel for whistleblowers to report trouble came with it too.
Charbel-Raphaël Ségerie runs the French safety group CeSIA. He signed the declaration himself, yet still wants a referee. "These decisions must rest with public authorities," Ségerie said8.
His own group read the text closely. It found the shared risk threshold missing. The deadline for developers to act stayed unset too.
Google Teaches AI to Prove Code Safe
DeepMind used teams of AI agents for a new kind of proof9. They wrote a mathematical proof for Signal's encryption code. WhatsApp and Google Messages lean on that same code.
The project is called AlphaProof Code. It produced more than 6.5 million lines of formal proof in four weeks. The proof confirmed four separate security properties, including whether a stolen phone can still expose yesterday's messages.
Swarat Chaudhuri works on the project as a DeepMind research scientist. He said the result opens a path more than it closes one. "We are very excited by this result, but there is a long distance to go before we are able to scale this to all the software of the world, and we want to do that," Chaudhuri said9.
Gemini's Priciest Model Opens to Everyone
Google widened access to Gemini 4 Carbon this week. The move ended a two-week trial limited to cybersecurity defenders10. Koray Kavukcuoglu runs AI architecture at Google DeepMind.
He called it the company's strongest model yet for long security reviews and multi-step agent work. The price holds at $2 per million input tokens. Output tokens cost $10 per million through Nov. 30, then the price doubles.
That introductory rate matches what Google charged for Gemini 4 Argon months earlier. The pattern is familiar enough now that developers can plan around it.
Musk Says Grok Runs on a New Engine
Elon Musk said xAI finished training Grok 4.8 this week. The model moved straight into reinforcement learning11. Engineers rewrote its entire software stack in C and C++ to get there.
They dropped the tools most labs use, JAX and SGLang. "We have dumped JAX/SGLang and rewritten the entire RL and inference stack from scratch in C/C++, which took some extra time. Training (aka pre-training) is done," Musk wrote.
A limited rollout to Grok's Bot users starts later this week. xAI built its own foundation instead of renting one. The wager pays off only if Grok's answers improve enough for people to notice.
The Justice Department Opens a New Front
A Super Intelligence Task Force opened at the Justice Department on Nov. 512. It absorbs an older AI enforcement unit. Attorney General Todd Blanche said the group exists to keep American labs ahead.
Catching bad actors who misuse the technology is the other half of its job. "SI development will continue to be powered by American innovators," Blanche said.
Prosecutors will also watch companies that oversell what their own systems can do. This same unit has aimed that focus at investors and customers before.
Companies Rehire the Staff AI Replaced
Nearly one in five U.S. workers let go because of AI have been asked to return13. The assessment firm SHL ran the survey this summer. Out of every 100 of those workers, 19 get the callback, and 81 keep job hunting on their own.
Returning workers want more than their old job back. Over half ask for a raise. Another 42 percent ask for a written promise their job stays safe.
Lucy Beaumont named the mistake plainly. She is SHL's global senior vice president of product and design. "Organizations are making fast, sometimes myopic, decisions about their people based on an incomplete picture of what their employees can do with or without AI," Beaumont said13.
Nearly 1 in 5 workers cut for AI get asked to come back
Share of U.S. workers replaced by AI who report being asked to return, one cube per percentage point
Each magenta cube is one percentage point of workers asked back; the gray cubes are the rest, who kept job hunting instead.
Source: Fortune [13]. Chart by Ryan Elliott Dennis.
The numbers behind this chart
| Item | Value |
|---|---|
| Asked to return | 19 |
| Kept job hunting | 81 |
| Whole | 100 |
Musicians Take Their AI Fight Back to the Table
The American Federation of Musicians returned to the bargaining table this month14. It is pressing Universal, Sony, Warner and Hollywood Records for a say over AI training. Its campaign is called "Human Made, Human Paid."
Consent comes first, before a recording trains a model. Credit comes next, when that training actually happens. A cut of what the deal pays rounds out the list. More than 80,000 musicians back the union's position heading into the session.
Tino Gagliardi framed the stakes plainly. He is the union's international president. "The record labels cannot pocket profits built on the backs of human artistry while cutting working musicians out of the equation," Gagliardi said14.
What to watch
Nov. 18 is the first test, when Anthropic prices its shares and the market says what today's demand was actually worth. Dec. 1 is the second. Gemini 4 Carbon's price doubles that day, and the number of developers still paying will tell Google who was just trying it out.
The Paris Call's own test arrives sooner: the next serious AI incident will show whether a shutdown mechanism actually gets used, or just signed.
Sources
- Ashley Capoot and Leslie Picker, "Anthropic sets IPO price range of $70 to $78, looks to start trading Nov. 19, sources say," CNBC, Nov. 10, 2026, https://www.cnbc.com/2026/11/10/anthropic-sets-price-range-of-70-to-78-in-upcoming-ipo.html
- Rebecca Bellan and Ram Iyer, "Anthropic sets IPO price range of $70 to $78 per share to reach an implied valuation of $1.404 trillion," TechCrunch, Nov. 10, 2026, https://techcrunch.com/2026/11/10/anthropic-sets-ipo-price-range-of-70-to-78-per-share-to-reach-an-implied-valuation-of-1-404-trillion/
- "Anthropic's IPO could be 'speculative fiction' or a 'generational moment' — investors are split," CNBC, Oct. 27, 2026, https://www.cnbc.com/2026/10/27/-anthropic-ipo-investors-growth-losses-s-1.html
- Kyle Guske II, "Anthropic Is The Most Ridiculous IPO of 2026," New Constructs, Nov. 6, 2026, https://www.newconstructs.com/anthropic-is-the-most-ridiculous-ipo-of-2026/
- Aisha Malik, "Anthropic launches Claude Meetings to turn workplace discussions into plans and tasks," TechCrunch, Nov. 10, 2026, https://techcrunch.com/2026/11/10/anthropic-launches-claude-meetings-to-turn-workplace-discussions-into-plans-and-tasks/
- Neetika Walter, "OpenAI launches Control Room with a powerful 35-second kill switch for rogue agents," Interesting Engineering, Nov. 9, 2026, https://interestingengineering.com/ai-robotics/openai-control-room-35-second-model-shutdown
- "Eight countries and top AI labs agree to shutdown mechanisms in new Paris declaration," The Decoder, Nov. 10, 2026, https://the-decoder.com/eight-countries-and-top-ai-labs-agree-to-shutdown-mechanisms-in-new-paris-declaration/
- "AI firms sign Macron's Paris Call, but safety group says it lacks teeth," TheNextWeb, Nov. 10, 2026, https://thenextweb.com/news/paris-call-ai-safety-openai-anthropic-cesia
- "Google DeepMind analysis proves Signal and WhatsApp encryption is secure," 9to5Mac, Nov. 10, 2026, https://9to5mac.com/2026/11/10/google-deepmind-analysis-proves-signal-and-whatsapp-encryption-is-secure/
- Ryan Merket, "Google makes Gemini 4 Carbon available at Argon's introductory price," RuntimeWire, Nov. 10, 2026, https://runtimewire.com/article/google-gemini-4-carbon-now-available
- "Grok 4.8 Is Coming This Week After a Full Rebuild, Says Musk," Tesla North, Nov. 10, 2026, https://teslanorth.com/2026/11/10/grok-4-8-coming-this-week-rebuild/
- Office of Public Affairs, "Department of Justice Launches Super Intelligence Task Force," U.S. Department of Justice, Nov. 5, 2026, https://www.justice.gov/opa/pr/department-justice-launches-super-intelligence-task-force
- "'AI rebounding': Nearly 20% of U.S. workers replaced by AI are being asked back," Fortune, Nov. 9, 2026, https://fortune.com/2026/11/09/ai-rebounding-20-us-workers-replaced-higher-pay-job-security/
- "'Human Made, Human Paid': The Largest U.S. Musicians Union Is Demanding AI Protections," Rolling Stone, Oct. 20, 2026, https://www.rollingstone.com/music/music-news/human-made-human-paid-musicians-union-afm-labor-agreement-1235645099/
