AI Daily: Lean Certifies a Claude-Aided Riemann Proof

The people who run Lean, the software mathematicians use to check proofs by computer, said on Saturday, 21 November 2026, that a proof of the Riemann Hypothesis found by an Anthropic mathematician working with Claude is correct. No human-readable paper has been published yet.

As we read it, the day’s coverage set a machine-checked result against open questions about money: a certified proof next to Oracle’s negative free cash flow, a tech sector at its widest gap below the S&P 500 in our data, and AI shopping agents spreading faster than their safeguards. Each item links to the full piece, which names its sources.

Business Pill · VERIFICATION COST

An evergreen Business Engineer explainer on verification cost, why work that is cheap to make is not cheap to check. It teaches the general idea only and does not discuss the stories in this roundup.

AI Daily Briefing · video

Today’s stories in about a minute (21 Nov 2026).

The key insight: As we read it, Saturday’s coverage set a machine-checked result against open questions about money: the Lean FRO and the Mathlib Initiative certified a Riemann proof found with Claude, while Oracle’s free cash flow was minus $5.4 billion last quarter and five of six AI cloud companies spent more than their operations brought in, per their 10-Qs and our arithmetic. We do not say the threads are linked.

Machines Checked the Maths

The Lean FRO and the Mathlib Initiative said in a joint statement on Saturday: “We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct.” The proof is by Levent Alpöge, a mathematician at Anthropic, who said on Friday he found it working with Claude. There is no human-readable paper yet, and the Clay Mathematics Institute’s prize rules require publication, two years and general acceptance. Read: the Lean statement.

The whole build checked by MIT’s Andrew Sutherland ran to 175,187 modules, about 21 times the 8,322 files it uses from Mathlib, by our arithmetic on his verification page. Alpöge and NYU mathematician Tristan Buckmaster also claim that the Navier-Stokes equations can break down in finite time with no outside force, in a paper dated 20 November, OfficeChai reported; that claim has not been certified. Read: the size of the build and the Navier-Stokes claim.

A paper co-written by Yann LeCun, posted on arXiv on 18 November, shows a world model’s score on one simulated robot “what if” task rising from 45.6% to 95.7%. MiniMax, the Chinese AI lab, says an internal-only model broke the designs of four candidate algorithms from past US NIST cryptography competitions, each in seconds, and that the work “does not affect deployed cryptographic standards”. Read: the LeCun paper and MiniMax’s claim.

Free cash flow in each company's latest quarter, operating cash flow minus capital spending, per its 10-Q: Ama
Free cash flow in each company’s latest quarter, operating cash flow minus capital spending, per its 10-Q: Amazon minus $34.8 billion, Alphabet minus $9.1 billion, CoreWeave minus $7.5 billion, Oracle minus $5.4 billion and Meta minus $1.2 billion, about $58.0 billion together; Microsoft had $18.6 billion left over. The figures and the sum are our arithmetic.

Who Pays for the Build

Oracle said it has “secured all the funding we need to fulfill our contract with OpenAI,” in a post on X on 20 November. Its free cash flow was minus $5.4 billion last quarter. Five of six AI cloud companies spent more on buildings and equipment than their operations brought in during their latest quarter, per their 10-Q filings, and only Microsoft had cash left over, $18.6 billion; the other five burned about $58.0 billion between them (our arithmetic).

Oracle spent about $1.47 on property and equipment for every $1 of revenue, against 43 cents at Microsoft (our arithmetic). Read: Oracle’s funding post and the cash ranking.

Lenders put about $48.2 billion into two AI borrowers on 19 and 20 November, while about $3.5 billion of other AI-linked debt stalled on weak demand in the same two days (our arithmetic). Apollo and Goldman Sachs priced roughly $44 billion of notes to buy Nvidia chips for SpaceX, and Reuters reported that “investor demand exceeded supply, allowing the transaction to be upsized.” Read: the lenders’ split.

Howard Marks asked Anthropic’s Claude to rate how predictable each part of the AI industry is, and it gave frontier labs such as OpenAI and Anthropic a 2 out of 10, in his 11 November memo, “A Layman Thinks About AI.” Read: the Marks memo.

Anthropic After the Listing

Before Anthropic priced its IPO at $83 a share, executives faced a Pentagon fight, a filing leak that panicked them and a Wednesday-night call over Gramercy Tavern takeout to raise the price, The Information reported. That $83 was $5, or 6.4%, above the top of the $70 to $78 range, by our arithmetic. Read: the inside story.

Anthropic left just over $20 billion on the table in its listing, the most in IPO history, IPO expert Jay Ritter told Fortune, about 37 times the roughly $550 million its banks earned (our arithmetic). Its shares changed hands 1,257,772,600 times in their first two days, 111.6% of the 1,127,000,000 shares sold, per Nasdaq data and our arithmetic. At most about 10.9% of the shares in our market-value math can trade today, and that can rise to about 20.8% by 2 February (our arithmetic from the final prospectus); both are upper limits. Read: the money left on the table, the trading volume and the float.

Markets and Borders

The tech sector fund XLK divided by the S&P 500 fund SPY closed on Friday, 20 November, 17.22% below its 2 June peak, by our arithmetic on Nasdaq.com closes; Wall St Engine called it the deepest relative drawdown since 2003. The PHLX Semiconductor index fell 17.1% in the week, and Taiwan will ban short selling below the previous day’s close from Monday 23 November, its Financial Supervisory Commission said. Read: the tech check, the chip index and Taiwan’s ban.

The Financial Times has traced an alleged route for moving Nvidia chips into China, The Next Web reported, summarising the FT investigation; Megaspeed’s Malaysian unit declared about $4.6 billion of imports from a renamed US arm of a blacklisted Chinese server maker in the year to December 2025, the FT reported. Nvidia said it put Megaspeed and a subsidiary on its block list in 2025, according to the report, and Megaspeed says it runs a lawful business. Read: the FT route.

Rhodium Group projects that 13 listed Chinese AI firms lift their AI infrastructure spending 103% this year, to 932 billion yuan ($139 billion), while China’s National Data Administration said it will improve standards for metering, billing and quality testing of AI tokens. Starting in January, Democratic governors will lead US states that hold 77% of the 300 gigawatts of data center capacity announced for 2027-2030, Rystad Energy said. Read: China’s spending, the token rules and the governors’ share.

Agents Start Shopping, and Leaking

Money spent by AI agents shopping through Stripe’s Link Agent Wallet was nearly 80 times higher in mid-November than at the start of September, Stripe said, giving no dollar figures. Meta’s Muse agent and Walmart “are about to launch the first agent-to-agent agentic commerce integration”, Marketplace Pulse’s Juozas Kaziukėnas wrote on X; neither company has announced a launch date. Read: Stripe’s data and Muse and Walmart.

Security researchers at NebuSec say one malicious webpage was enough to take over Muse and send a user’s private chats and stored memories to an outside server; NebuSec says Meta had patched the exploits by the time it published. OpenRouter told some customers that prompts they sent to MiniMax models “may have been read” by a third party, and OpenAI removed a sentence from its ChatGPT shopping help page that said product results were not influenced by its partnerships. Read: the Muse hijack, OpenRouter’s email and the shopping page change.

Bar chart: the Lean build certified on 21 November 2026 ran to 175,187 modules against the 8,322 Mathlib files it uses, about 21 times
Size of the Riemann proof build checked by MIT’s Andrew Sutherland: 175,187 modules, about 21 times the 8,322 Mathlib files it uses; the ratio is our arithmetic on his verification page, as our piece reports.

The Structural Read

As we read it, the proof thread is about who checks the work: the Lean FRO and the Mathlib Initiative certified both the statement and the proof, while the Clay Mathematics Institute’s prize rules still require publication, two years and general acceptance, as our piece reports.

As we read it, the money thread is about who carries the build: lenders put about $48.2 billion into two AI borrowers on 19 and 20 November while about $3.5 billion of other AI-linked debt stalled, by our arithmetic, and Oracle spent about $1.47 on property and equipment for every $1 of revenue, by our arithmetic.

As we read it, the agent thread is about reach before safeguards: Stripe said agent spending through its wallet was nearly 80 times higher than in early September, while NebuSec says one malicious webpage was enough to take over Meta’s Muse agent before Meta patched it, as NebuSec reports.

The Lean FRO and the Mathlib Initiative, in a joint statement posted on Saturday and quoted in our piece

“We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct.”

Three Implications

READERS OF AI CLAIMS The Riemann proof is certified by Lean, but there is no human-readable paper yet, and the Navier-Stokes and MiniMax claims have not been certified, as our pieces report.

LENDERS AND SHAREHOLDERS Five of six AI cloud companies spent more on buildings and equipment than their operations brought in during their latest quarter, per their 10-Q filings and our arithmetic; this is not investment advice.

SHOPS AND SHOPPERS Neither Meta nor Walmart has announced a launch date for the Muse integration, and OpenAI has not publicly explained the change to its shopping page, as our pieces report.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Supercycle.

The analysis says of its frame: “the term supercycle should not be interpreted as a purely economic concept. It goes far beyond economics.”

As we read it, Saturday touched more than the markets: a US chip rule hitting data centers in Malaysia, Chinese standards for billing AI tokens, a Taiwanese short-selling ban and a mathematics result checked by software. We do not say the analysis predicted these items. It was published in June 2026 and does not discuss them.

What Is Not Established

This roundup adds no new reporting and selects from the day’s coverage. Every figure above is taken from the linked FourWeekMBA piece, each built from its own primary source. The threads, and the reading of the day as a checked result set against open questions about money, are our grouping, not a finding.

We do not claim that any one story moved prices or caused another. The Riemann proof is certified by Lean but not yet published or accepted under the Clay rules; the Navier-Stokes and MiniMax claims are unverified; the FT route and The Information’s account rest on their reporting; the Muse and Walmart launch is unannounced; the market figures are our arithmetic; and nothing here is a forecast.

This Week In The AI Supercycle

A World Without an Order, and an AI Central Bank Without a Successor

This week’s issue: the week AI memory stayed largely sold out while the people holding the shares paid: why the AI Supercycle is not a dot-com replay, read through geopolitics, macro and technology, with the gap between building and financing the widest we’ve recorded.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer on how geopolitics, culture and the economy connect in the AI era.

Read the analysis →

Get the premium package at 50% off →

The Bottom Line

For anyone pricing the AI build, Saturday set Lean’s certification of a Claude-aided Riemann proof next to Oracle’s minus $5.4 billion of free cash flow and a tech sector 17.22% below its June peak against the S&P 500, by our arithmetic.

For the frame behind the day, read This Week in the AI Supercycle on The Business Engineer: read the analysis.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. This roundup draws on FourWeekMBA’s news coverage of 20 and 21 November 2026, linked above; every story was built from its own primary source, which it cites and links. We re-read each piece on 21 November 2026 and took every figure here from it. The threads and the figures in our charts are our own grouping and selection. Nothing here is a forecast, and nothing here is financial or investment advice.

Sources: FourWeekMBA: Lean Certifies Claude-Aided Riemann Proof, 167 Years On · FourWeekMBA: Riemann Proof Build: 175,187 Modules, ~21x Its Mathlib Files · FourWeekMBA: 2nd Millennium Claim in a Day: Unforced Navier-Stokes Blowup · FourWeekMBA: LeCun Paper Lifts Robot ‘What If’ Scores From 46% to 96% · FourWeekMBA: MiniMax Says Its AI Broke 4 Old NIST Crypto Candidates · FourWeekMBA: Oracle Says OpenAI Deal Fully Funded; $116 Shares Now $101 · FourWeekMBA: 5 of 6 AI Clouds Burned Cash; Oracle Spent $1.47 per $1 Sold · FourWeekMBA: AI Lenders Split: $48.2B Priced in 2 Days, $3.5B Stalled · FourWeekMBA: Howard Marks: Claude Rates Frontier AI Labs 2 Out of 10 · FourWeekMBA: Inside Anthropic’s $83 IPO: A Pentagon Fight, Report Says · FourWeekMBA: Anthropic Left a Record $20B on the Table, 37x Its Bank Fees · FourWeekMBA: 1.26B Anthropic Shares Traded in 2 Days, 112% of the IPO · FourWeekMBA: 11% of Anthropic’s Shares Can Trade Now, Up to 21% by Feb. 2 · FourWeekMBA: Tech Is 17.2% Below Its High vs the S&P 500: We Checked · FourWeekMBA: Chip Index Falls 17.1% in a Week, 42.5% Below Its June High · FourWeekMBA: Taiwan Bans Short Sales After a 17.5% Weekly Stock Plunge · FourWeekMBA: FT Maps $4.6B Chip Route; Nvidia Says It Blocked Megaspeed · FourWeekMBA: China’s AI Capex Doubles to $139B; Its AI Labs Earn 7.7% · FourWeekMBA: China Plans AI Token Billing Rules as Daily Use Tops 500T · FourWeekMBA: Democratic Governors to Lead 77% of US Data Center Pipeline · FourWeekMBA: Stripe: AI Agents’ Spending Up Nearly 80x Since September · FourWeekMBA: Muse to Shop Walmart and Query Its AI Sparky, Analyst Says · FourWeekMBA: 1 Webpage Let Researchers Pull Meta Muse Chats and Memories · FourWeekMBA: OpenRouter: Test Links Sent to MiniMax Opened 40-130h Later · FourWeekMBA: OpenAI Drops ChatGPT Shopping Line on Partnership Influence

Scroll to Top
FourWeekMBA