Sat, Nov 21, 2026Saturday, November 21, 2026 · 20 stories · 7 min read

Anthropic's first public week ends with a Claude Riemann proof claim

Ramit KoulFounder, Software Engineer & Innovator · Published 5:04 AM ET

A day after Anthropic's stock market debut, one of its mathematicians, Levent Alpöge, said he and Claude had proved the Riemann hypothesis, and mathematicians are now checking the claim.1, 2 On the same Friday, President Trump threatened a "tremendous tax" on US companies that build AI data centers abroad, and AI stocks kept falling.3, 4 A record IPO, a claimed breakthrough in mathematics and a vague threat to infrastructure plans all landed in the same week. Here are the 3 stories that matter most, then 17 briefs. Numbers in the text link to the references at the end.

Research

Anthropic mathematician says Claude helped prove the Riemann hypothesis

Image: scientificamerican.com

Levent Alpöge, a mathematician at Anthropic, posted a four-page note on his website Friday saying he believes that, working with Claude, he has found a proof of the Riemann hypothesis.5, 6 He says the method also resolves the Generalized Riemann Hypothesis, and he expects it to reach the Extended Riemann Hypothesis with more time.7 No outside experts have reviewed the claim yet.5

Alpöge released only a version of the proof written in Lean, a language that lets a computer check every step of an argument.2 New Scientist reports the Lean code runs to 17 million lines, and Alpöge says an unpublished human-readable version ran past 250,000 pages.8 He called posting that version an "unreadable travesty" that would be inappropriate to share.2 Kevin Buzzard of Imperial College London said the first step is to see whether the code compiles.2 Thomas Bloom of the University of Manchester warned that the AI could be exploiting an undiscovered bug even if it does.8

Alpöge says he told CEO Dario Amodei about the result and held the announcement until after Thursday's IPO.7, 2 Anthropic had not commented as of Friday.7 To win the Clay Mathematics Institute's $1 million prize, a solution must appear in a peer-reviewed journal and withstand two years of scrutiny.5 Separately, the president of the American Mathematical Society, Ravi Vakil, pointed to a 940-page proof of the generalized hypothesis by Alessandro Masullo, written without AI.8

Why it matters for builders

If the Lean proof holds up, formal verification will have checked a result that no human has fully read, and builders in other fields will face the same question of how to trust AI output nobody can follow.8 The checking process over the coming weeks will show how robust today's proof tooling really is.2

Policy & legal

Trump threatens a tremendous tax on US firms building data centers abroad

President Trump said Friday that US companies will not be allowed to build data centers outside the country and that any that try will face "tremendous penalties".3 He also said companies could still build abroad but would pay "a tremendous tax" for doing so.9 He gave no details and said he had not yet told his own staff about the plan.10

Trump framed the plan around China, saying China "will take all of them" if data centers go overseas.11 Axios noted that Congress, not the president, holds the constitutional power to impose federal taxes.11 SiliconANGLE pointed out that it is unclear whether the threat covers only AI-optimized sites or also general cloud regions.12 That uncertainty matters because AWS plans to invest 18 billion euros in Germany and Google has committed 13 billion euros in Finland.12

The remarks come as voters push back on data centers at home: local ballot measures in the midterms largely moved to restrict them.13 Meta's planned site near Edmonton and the Nova Scotia data center that Anthropic has leased for Claude are among US projects in Canada.9 The Senate passed a permitting bill on Thursday that would speed up energy projects, and Trump urged the House to pass it.10

Why it matters for builders: Teams that rely on overseas regions for latency, redundancy or data residency should watch for any concrete policy, since those needs often require infrastructure abroad.11 For now this is a verbal threat with no mechanism behind it.10

Funding & deals

Anthropic closes its first trading day up 17% as markets slide

Anthropic's shares closed their first day on the Nasdaq about 17% above the $83 IPO price, after briefly valuing the company above $2 trillion early in the session.14 Fortune put the closing valuation at $1.27 trillion, or $1.76 trillion depending on the method.14 The debut came on a down day for the market: the Nasdaq fell 2.25% and the Dow fell 2.19%.14

The sell-off continued Friday. The S&P 500 dropped 3.7% and entered a correction, and the Nasdaq composite turned negative for the year.4 NBC News reported that the record IPO did little to ease investor worries about AI stocks.3

According to PitchBook, Anthropic has more than 400 venture backers and none holds more than 5%.14 Investors quoted by Fortune said the listing will force more scrutiny of inference costs, capital efficiency and unit economics across AI companies.14

Why it matters for builders: Anthropic's revenue from Claude Code, subscriptions and enterprise deals is now a matter of public record, so pricing and capacity decisions that affect API users will play out in front of investors.14

In brief

  • Models

    Claude now hides a watermark in the text it writes. According to an O'Reilly Radar analysis, Anthropic's support page says its newest models and some older ones watermark text on every Claude surface worldwide, a step tied to the EU AI Act's machine-readable marking requirement.15 Neither Anthropic, OpenAI nor Google lets the public check text against its detector.15

  • Models

    Baseten released bison-1 and groundhog-1, open decision models built on Gemma 4. Baseten says bison-1 ranks first of 19 models on the Artificial Analysis Decision Index v2.1 with a score of 76. Both models take text and images with a 128k context and cost $0.10 and $0.035 per million tokens.16, 17 Requests longer than the context limit are refused rather than truncated, so pipelines need a length check.16

  • Industry

    Nvidia's interim CEO Colette Kress put new limits on financing for AI cloud partners. On her first earnings call as interim CEO, Kress reported a $115.8 billion quarter. She said future help for partners will focus on land, power and shell, and that commitments to neoclouds apply only if operators secure their own financing and deploy the capacity.18 Nvidia's maximum exposure across infrastructure guarantees reached $118 billion at the end of the October quarter.18

  • Policy & legal

    Cursor is weighing a move away from Chinese open-weight models, a House report says. A joint report from two House committees says the SpaceX-owned coding tool is considering phasing out Chinese models, and notes that its Composer 2 was built on Moonshot AI's Kimi K2.5.19 The report recommends banning models developed in China from federal systems and federal contract work.19

  • Models

    RadixArk released DaisyRose-27B, an agentic coding model trained with reinforcement learning only. The Apache 2.0 model is post-trained from Qwen3.8-27B and more than doubles its Terminal Bench 4.1 score, from 5.6 to 12.0. Training ran 18 days on 144 H200 GPUs.20 It fits on one GPU, at about 52 GB in BF16 and 23 GB in NVFP4, and all reported scores are self-reported.20

  • Industry

    Signal now blocks screenshots by default to keep OS-level AI agents out of chats. Signal's new Block Siri AI setting on iOS 27 is on by default, and on Android the app turns on Screen security when it detects an assistant reading the screen.21, 22 President Meredith Whittaker called the fix incomplete, said Signal cannot do the same on macOS, and asked Apple, Google and Microsoft to give developers a documented way to opt out of AI data access.21, 23

  • Models

    Nvidia released Nemotron-Labs-Sigma, a family of continuous diffusion language models. The 14 open checkpoints at 3B and 8B scale denoise text in a 16-dimensional embedding space. Nvidia reports that the 8-step distilled 8B model keeps a 70.34 average on its six-task suite.24 The self-speculative variants ran 3.46 times faster than autoregressive decoding on a GB200 with SGLang, by Nvidia's own measurements.24

  • Policy & legal

    House data shows US officials approved nearly 900,000 AI chips for Chinese companies. A Commerce Department summary dated May 11 and released by the House Foreign Affairs Committee shows 879,868 Nvidia and AMD chips approved out of applications for 3.44 million. Approved buyers included Tencent, Alibaba and Baidu.25 The Washington Post reported that almost 580,000 of the approved chips were Nvidia's, and Nvidia said its H200 shipments were legal and authorized.26, 25

  • Products

    OpenAI brought ChatGPT Space to Plus subscribers. Space is a workspace for documents, slides, spreadsheets and sites, and it can automatically refresh pages using connected apps such as Gmail and Slack on a schedule.27 It was previously limited to Pro, Business and Enterprise plans.27

  • Products

    WiCi released a free desktop app that runs 800B-parameter models by streaming weights from the SSD. WiCi Infer is built on the open-source InferSSD engine, a fork of llama.cpp. It keeps routed expert weights on disk, needs about 520GB of free SSD space and at least 16GB to 24GB of memory.28 WiCi's own benchmarks show about 2.7 tokens per second for DeepSeek V4.1 Flash on a 48GB M5 Max MacBook Pro.28

  • Funding & deals

    After its IPO, Anthropic faces the question of what follows Claude Code. Business Insider reports that Anthropic, which closed its first trading day valued at $1.7 trillion, now needs a business beyond AI coding, and points to biotech, cybersecurity and robotics as possible areas.29

  • Research

    METR updated its time horizon chart and found models still fall short on judgment-heavy tasks. METR says time horizons on its combined task suites are doubling roughly every 3.6 months.30 When asked to design AI evaluations, though, no model produced one that METR staff found useful for their research.31 Newer models such as Opus 5.5 reached what METR calls borderline above the bar on that task.31

  • Products

    OpenAI is testing a redesigned ChatGPT mobile app with a dedicated tab for dots. The new layout adds a Home tab, a Dot tab and a grid of tools such as Codex, Space and Projects that users can pin to the tab bar. It is rolling out first to eligible Pro users, outside the EEA, the UK, Switzerland and South Korea.32, 33

  • Industry

    A hosting breach aimed at AI provider keys is pushing firms to treat agent keys like user logins. An Accounting Today column describes a cloud hosting platform disclosing that attackers sought AI provider API keys, and urges firms to scope, rotate and log each agent's keys under the FTC Safeguards Rule.34

  • Products

    Google's Gemini can now generate GIFs. The looping animations run about three seconds and use the Gemini Omni 1.2 Flash video model, initially in seven countries including the US, UK and Japan.35

  • Models

    OpenAI introduced MedChemBench to test AI reasoning in medicinal chemistry. The benchmark has 99 problems drawn from 117 papers and patents, and GPT-6.2 Astra scored 84.5%.36, 37 OpenAI says the results do not show that models can discover drugs on their own.36

  • Infrastructure

    A lawyer's $9,450 server built from 12 used V100s now runs every DeepSeek V4.1 Flash expert at NVFP4. With a third board of cards, all 15,360 experts fit on the GPUs, and the time to the first code token at 256K context fell from 108.4 seconds to 58 seconds.38

References

Every source behind this edition. Open one to read the full story.

  1. 1Anthropic mathematician claims to prove Riemann hypothesis, most famous problem in all of mathscientificamerican.com · scientificamerican.com
  2. 2Anthropic mathematician claims AI-assisted proof of the Riemann hypothesisnewscientist.com · newscientist.com
  3. 3Trump threatens ‘tremendous penalties’ on companies that build AI data centers overseasNBC News · nbcnews.com
  4. 4AP News in Brief at 11:04 p.m. ESTphillytrib.com · phillytrib.com
  5. 5Anthropic researcher says AI chatbot Claude solved math’s greatest riddlestraitstimes.com · straitstimes.com
  6. 6Anthropic Mathematician Claims to Have Solved Math’s Most Famous ProblemThe New York Times · nytimes.com
  7. 7Anthropic mathematician claims to have solved prestigious Millennium Prize problem with ClaudeThe Verge · theverge.com
  8. 8Mathematicians are amazed by ‘holy grail’ proof of Riemann hypothesisnewscientist.com · newscientist.com
  9. 9Trump threatens U.S. firms building data centres elsewhere with major taxesglobalnews.ca · globalnews.ca
  10. 10US tech companies that build AI data centers in other countries will be penalized, Trump saysAP News · apnews.com
  11. 11Trump floats taxing U.S. companies constructing data centers overseasAxios · axios.com
  12. 12Trump announces plan to tax US companies that build data centers abroad - SiliconANGLESiliconANGLE · siliconangle.com
  13. 13Trump threatens to tax US companies that build data centers overseasmiamiherald.com · miamiherald.com
  14. 14Anthropic is now AI’s bellwetherFortune · fortune.com
  15. 15Claude Watermarks Your Words, and You Can’t Check Them (Yet)oreilly.com · oreilly.com
  16. 16Baseten Releases Bison-1 and Groundhog-1, Topping Decision Model Indexmarktechpost.com · marktechpost.com
  17. 17Meet bison-1 and groundhog-1, frontier American decision modelsbaseten.co · baseten.co
  18. 18Nvidia’s Colette Kress Signals the End of No-Strings A.I. Financingobserver.com · observer.com
  19. 19SpaceX-owned Cursor weighs phasing out Chinese AI modelstechinasia.com · techinasia.com
  20. 20RadixArk Ships DaisyRose-27B, a 27B Agentic Coding Model Trained With Reinforcement Learning Onlymarktechpost.com · marktechpost.com
  21. 21Signal locks out Siri AI and Gemini, but says it cannot protect the MacThe Next Web · thenextweb.com
  22. 22Signal blocking screenshots to keep AI assistants out of private chatsMashable · mashable.com
  23. 23Signal Is Blocking Screenshots to Protect Your Messages from AI AgentsCNET · cnet.com
  24. 24NVIDIA AI Releases Nemotron-Labs-Sigma, a Continuous Diffusion Language Model with 14 checkpoints at 3B/8Bmarktechpost.com · marktechpost.com
  25. 25Trump Admin Approved Nearly 900,000 AI Chips For China, House Data Shows: Nvidia Says H200 Shipments Were ‘Legal And Authorized’Benzinga · benzinga.com
  26. 26Officials approved the sale of 580,000 Nvidia chips to Chinese companieswashingtonpost.com · washingtonpost.com
  27. 27ChatGPT just got a massive upgrade — it can now create documents that update themselvestomsguide.com · tomsguide.com
  28. 28WiCi Infer Runs 800B-Parameter AI Models on a Laptop by Streaming Them From the SSDhackernoon.com · hackernoon.com
  29. 29After its blockbuster IPO, Anthropic's next challenge is finding a hit bigger than Claude CodeBusiness Insider · businessinsider.com
  30. 30AI R&D-relevant model capabilities are improving rapidly and are getting harder to measuremetr.org · metr.org
  31. 31The most misunderstood graph in AI just got a major updateMIT Technology Review · technologyreview.com
  32. 32ChatGPT is getting a new look on your phone, and your dot now has its own tabDigital Trends · digitaltrends.com
  33. 33OpenAI’s ChatGPT app redesign adds a dedicated Dots tabandroidauthority.com · androidauthority.com
  34. 34Your firm's AI keys reach client data. Put them in your WISPaccountingtoday.com · accountingtoday.com
  35. 35Gemini can now be your go-to source for making GIFsandroidauthority.com · androidauthority.com
  36. 36Can AI Help Find Blockbuster Drugs? OpenAI Tests It OutBenzinga · benzinga.com
  37. 37Can AI help scientists identify promising drug candidates?OpenAI · openai.com
  38. 38Lawyer’s $9,450 private AI server with 12 used Nvidia V100s now runs every DeepSeek V4.1 Flash expertTom's Hardware · tomshardware.com

That’s the edition.

The next one is ready tomorrow by 6 AM ET.

Written with AI from the sources in the references above.

Ramit Koul
AuthorRamit KoulFounder, Software Engineer & Innovator