Top AI Stories · November 21, 2026

A Claude-aided Riemann proof claim; AI stocks slide

An Anthropic mathematician posted a proof of the Riemann hypothesis he says he found with Claude, and mathematicians are still checking it. Separately, Anthropic's shares fell 9 percent as AI stocks sold off. Plus 6 more stories.

Listen to this brief

Free preview · first 0:30
0:00 / 0:30

Audio & video are paid features

Plus unlocks audio streaming and PDF downloads. Pro adds offline MP3 downloads, video, certificates, and more.

Plus adds:
  • Audio streaming
  • Downloadable PDFs
  • All AI Playbooks
  • Personalized content
Pro also adds:
  • Certificates of completion
  • Audio MP3 downloads
  • Video lessonssoon
  • & More…soon

An Anthropic mathematician posted a computer-checked proof of the Riemann hypothesis that he says he found with Claude, and mathematicians have started the long work of understanding it. Separately, Anthropic's shares gave back most of their first-day gain as a broad sell-off hit AI stocks.

  1. 1

    An Anthropic mathematician says he proved the Riemann hypothesis with Claude

    Levent Alpöge, a mathematician at Anthropic, posted a proof of the Riemann hypothesis, the problem from 1859 about how prime numbers are distributed, and says he found it working with Claude. He says the method also settles the Generalized Riemann Hypothesis. He released it only in Lean, a proof-checking language, roughly 17.4 million lines by Scott Aaronson's count, because no readable write-up exists yet. Mathematicians including MIT's Andrew Sutherland report that it compiles, but no one has yet explained the argument, and Terence Tao expects digesting it to take time. In August, Anthropic said it did not expect Claude's methods to lead to a proof. The same day, Stanford's Ravi Vakil posted an all-human proof of the generalized version by postdoc Alessandro Masullo, the product of 12 years of work and not yet formalized.

  2. 2

    Anthropic shares fall 9 percent as the Nasdaq wipes out its gains for the year

    Anthropic's stock fell 9 percent on its second day of trading, to about $88, close to its $83 IPO price, after rising 17 percent on its first day. The selling ran across AI-linked stocks: Nvidia and SanDisk each fell 14 percent and Micron 12 percent, and the Nasdaq Composite closed down 5 percent for 2026. Taiwan's market dropped 6.7 percent and SoftBank had its worst day in more than two years. One analyst at the trading platform IG suggested investors were selling other tech shares to pay for Anthropic stock.

  3. 3

    Two million planted secrets show AI prompts leaking across the resale chain

    HD Moore, chief research officer at the security firm Dragos, sent nearly two million prompts carrying unique, never-used secrets, such as cloud access keys and private web addresses, through AI providers and routers from July to November, then watched where they turned up. Attackers had breached at least three hosting providers and copied requests at the front door, including ones marked zero data retention. Some cheap resellers logged prompts, and keys sent through them were later used. A Chinese AI lab's systems visited the secret links in more than half the prompts sent to its models, even when they went through US hosts.

  4. 4

    Trump threatens a 'tremendous tax' on companies that build data centers abroad

    Speaking after his summit with China's Xi Jinping and Russia's Vladimir Putin, President Trump said the US would not let its companies build data centers in other countries, and that any that did would face a "tremendous tax." He gave no details and said he had not yet raised the idea with his own staff. The threat comes as states and cities pass moratoriums and new limits on data center projects, which could push builders overseas, and a few months before a Democratic majority takes over Congress.

  5. 5

    OpenAI says it cut a reasoning-theft attack from 54 percent success to 0.2 percent

    OpenAI described how it fights adversarial distillation, where someone trains a rival model on another model's outputs. It concentrates on the hidden reasoning its models produce, and says an internal red-team attack that recovered that reasoning 54 percent of the time against GPT-6 Astra succeeded 0.2 percent of the time against GPT-6.2 Astra. Its weekly enforcement actions have hit thousands to tens of thousands of accounts, rising over the summer. OpenAI again linked a September campaign to people tied to Moonshot AI, promised more on recent campaigns, and backed the Chip Security Act among other bills.

  6. 6

    Meta reconsiders its 2030 net-zero goal as its emissions jump nearly 40 percent

    Meta's 2026 sustainability report says it is re-evaluating the 2030 deadline for its net-zero goal because demand for AI infrastructure grew faster than almost anyone projected. Its total emissions rose from 15.6 million to more than 21.7 million metric tonnes of carbon dioxide equivalent in 2025, the steepest rise reported by any of the four biggest US tech companies. Microsoft reported a 25 percent rise, Google 18 percent and Amazon 16 percent, and all three have kept their goals so far. Meta's water use in water-stressed areas also climbed, from 198 million to 246 million gallons.

  7. 7

    Signal blocks screenshots on phones to keep AI assistants out of chats

    Signal now blocks screenshots on iOS 27, and on Android devices where it detects an AI assistant reading the screen, to stop built-in agents such as Siri and Gemini from capturing messages. President Meredith Whittaker says Apple, Google and Microsoft give app makers no clean way to wall off their data from these agents, which is why Signal cannot do the same on macOS yet. Users can switch screenshots back on, and she calls the fix incomplete, since agents may reach app data in other ways, such as through the file system. Signal used the same approach against Microsoft's Recall on Windows.

  8. 8

    Thinking Machines releases Inkling 2 Mini, an open model small enough for a laptop

    Thinking Machines Lab's new model runs its middle layers three times for every token, a looped design meant to reason more without writing more. It is a mixture-of-experts (MoE) model with 26.4 billion total parameters and 3.4 billion active, and runs on a Mac with at least 32 GB of memory through Ollama. The weights are on Hugging Face under Apache 2.0. The lab says it leads open models under 100 billion parameters on the DeepSWE coding benchmark, but trails Alibaba's 27-billion-parameter Qwen model on multi-turn chat and complex instructions. It arrives about a week after the lab published the weights for its larger Inkling 2 Small.

Get Top AI Stories by email

The day's most important AI news — free, daily, unsubscribe anytime.

Share
Spot a typo or have feedback?Share feedback

Sources

  1. 1.Trump threatens 'tremendous tax' for moving AI data centers outside the US — The Verge · November 20, 2026
  2. 2.Signal is blocking Gemini and Siri AI from reading your screen — The Verge · November 20, 2026
  3. 3.Introducing Inkling 2 Mini — Thinking Machines Lab · November 20, 2026
  4. 4.Meta reconsiders 2030 net-zero goal amid record emissions from AI data centers — Ars Technica · November 20, 2026
  5. 5.The stochastic parrots have apparently proved the Riemann Hypothesis — Shtetl-Optimized · November 20, 2026
  6. 6.Unforced Euler singularity, unforced Navier-Stokes singularity, and the Generalized Riemann Hypothesis — Terence Tao · November 20, 2026
  7. 7.Anthropic mathematician claims to have solved prestigious Millennium Prize problem with Claude — The Verge · November 20, 2026
  8. 8.Nasdaq erases all 2026 gains as AI sell-off dampens Anthropic's second day of trading — The Guardian · November 20, 2026
  9. 9.Signal is Serious About Protecting Your Privacy — Signal · November 20, 2026
  10. 10.Researcher plants 2m canaries into the AI supply chain. A privacy disaster surfaced. — Ars Technica · November 20, 2026
  11. 11.Announcement of a proof of the Riemann Hypothesis (PDF) — Levent Alpöge · November 20, 2026
  12. 12.AI claimed a solution to the 167-year-old Riemann hypothesis — and a human did, too — TechCrunch · November 20, 2026
  13. 13.Combatting adversarial distillation — OpenAI · November 20, 2026
  14. 14.AI Supply Chain Security: What Canary Tokens Found — Dragos · November 19, 2026

AI disclosure: Researched and drafted with AI; reviewed and edited by the AI Pro Playbook editorial team before publishing. Sources above link to original publishers.

🧭Recommended for you