Skip to content

AI News Nov 21, 2026

Language: DE / EN / ZH / FR / ES / PT / JA / KO

AI-generated: summaries written by AI from the linked sources. How we use AI

Key Takeaways

  • Anthropic mathematician Levent Alpöge says he used Claude to produce a proof of the Riemann hypothesis, formalized in Lean and spanning over 250,000 pages. Hours later, Stanford re…
  • NYU's Tristan Buckmaster and Anthropic's Levent Alpöge published an AI-aided proof of Navier-Stokes blowup, saying it closes a loophole in OpenAI's earlier solution. The work comes…
  • Researchers planted 2 million canary credentials in requests sent through AI routers including OpenRouter and Vercel, and found that privacy protections were often broken and keys …
  • Databricks: several billion dollars (Unknown)
  • Amazon → Decart ($7.5B)
  • Use OpenAI's published prompt templates for GPT-6.2 Astra and Sol to address overly brief answers, unnecessary follow-up questions, and excessive testing when c…

Why Today Matters

  • AI math proofsTwo AI-assisted math claims arrived the same day: Levent Alpöge says Claude produced a 250,000-page Lean proof of the Riemann hypothesis, while Stanford researchers published a human competing proof. Mathematicians are still checking whether either holds up, so neither is established.
  • AI research raceAnthropic-linked proof work extends to physics-adjacent problems, with Buckmaster and Alpöge publishing an AI-aided Navier-Stokes blowup proof they say closes a loophole in OpenAI's earlier solution, two months after being scooped by OpenAI.
  • AI security and policySecurity and geopolitics are converging on AI infrastructure: researchers found canary credentials stolen through routers like OpenRouter and Vercel despite zero-data-retention policies, while Cursor told lawmakers it is considering phasing out Chinese open-weight models such as Kimi K2.5.
  • Efficiency vs scaleEfficiency claims are mounting against scale: Nvidia researchers argue large context windows are often slow and error-prone, and Thinking Machines' Inkling 2 Mini reportedly ties Qwen3.8 27B on Artificial Analysis's Intelligence Index at roughly six times lower cost, per the company.
  • AI funding and M&ACapital keeps concentrating: Databricks is reportedly in early talks to raise several billion dollars at about a $225B valuation, up from $190B in August, while Amazon is reportedly nearing a $7.5B deal for Decart to help AWS compete with Nvidia.

AI-generated analysis — learn how we work

What are the top AI breakthroughs?

This Nov 21, 2026 covers 10 curated AI news items spanning technology, research, and product developments. Anthropic mathematician Levent Alpöge says he used Claude to produce a proof of the Riemann hypothesis, formalized in Le...

Anthropic mathematician Levent Alpöge says he used Claude to produce a proof of the Riemann hypothes…

Anthropic mathematician Levent Alpöge says he used Claude to produce a proof of the Riemann hypothesis, formalized in Lean and spanning over 250,000 pages. Hours later, Stanford researchers published an independent, entirely human competing proof, and mathematicians are now checking whether either holds up.

Category: Research|Impact:critical|Source: The Decoder|Read brief

NYU's Tristan Buckmaster and Anthropic's Levent Alpöge published an AI-aided proof of Navier-Stokes …

NYU's Tristan Buckmaster and Anthropic's Levent Alpöge published an AI-aided proof of Navier-Stokes blowup, saying it closes a loophole in OpenAI's earlier solution. The work comes two months after the pair were scooped by OpenAI on the problem.

Category: Research|Impact:high|Source: Techmeme|Read brief

Researchers planted 2 million canary credentials in requests sent through AI routers including OpenR…

Researchers planted 2 million canary credentials in requests sent through AI routers including OpenRouter and Vercel, and found that privacy protections were often broken and keys stolen, even under zero-data-retention policies.

Category: Safety|Impact:high|Source: Techmeme|Read brief

Bytedance and several US universities introduced Axion 1.0, an AI agent that the researchers say riv…

Bytedance and several US universities introduced Axion 1.0, an AI agent that the researchers say rivals top commercial models from OpenAI and Anthropic on benchmarks after just three training phases. They attribute the gains to curated training data rather than more training steps, and released the model, data, and code freely.

Category: AI Models|Impact:high|Source: The Decoder|Read brief

Nvidia researchers argue that large context windows are often unnecessary, being slow, expensive, an…

Nvidia researchers argue that large context windows are often unnecessary, being slow, expensive, and error-prone. In tests, periodic summaries and targeted keyword search usually matched full-context results at a fraction of the cost, and the team proposes a new class of models that works with short notes instead of long context.

Category: Research|Impact:high|Source: The Decoder|Read brief

Thinking Machines released Inkling 2 Mini, an open looped transformer that runs 12 of its 24 layers …

Thinking Machines released Inkling 2 Mini, an open looped transformer that runs 12 of its 24 layers three times per token, letting it reason internally for longer without generating extra text. On Artificial Analysis's Intelligence Index it ties with Qwen3.8 27B at roughly six times lower cost, and it runs on a laptop.

Category: AI Models|Impact:high|Source: The Decoder|Read brief

Cursor told lawmakers it is considering phasing out Chinese open-weight AI models due to geopolitica…

Cursor told lawmakers it is considering phasing out Chinese open-weight AI models due to geopolitical risk, according to CNBC. Cursor's Composer 2 was built on Kimi K2.5.

Category: Industry|Impact:high|Source: Techmeme|Read brief

Signal is restricting users' ability to take screenshots in both its Android and iOS apps, as the co…

Signal is restricting users' ability to take screenshots in both its Android and iOS apps, as the company confronts what it calls the existential threat of AI surveillance.

Category: Safety|Impact:medium|Source: TechCrunch AI|Read brief

UN diplomats meeting in Geneva failed to agree on rules regulating autonomous weapons, or so-called …

UN diplomats meeting in Geneva failed to agree on rules regulating autonomous weapons, or so-called killer robots. Representatives left agreeing only to start discussing options after a largely fruitless week of negotiations.

Category: Policy|Impact:medium|Source: The Verge AI|Read brief

Tencent announced a public beta of WorkBuddy's Computer Use feature, which targets desktop apps that…

Tencent announced a public beta of WorkBuddy's Computer Use feature, which targets desktop apps that lack APIs. The beta comes with client-version requirements, system permission needs, and supervision cautions.

Category: Product|Impact:medium|Source: Pandaily|Read brief

Top AI Videos This Week

2 curated YouTube videos about AI developments.

Based on the publisher's description: BBC News examines self-improving AI, which researchers say is …

Based on the publisher's description: BBC News examines self-improving AI, which researchers say is …

Based on the publisher's description: BBC News examines self-improving AI, which researchers say is moving from science fiction to serious study. The video explores whether such systems could accelerate scientific discovery or improve themselves faster than humans can control, presenting both optimistic and cautionary perspectives.

Read brief

Based on the publisher's description: Dan Niles of Niles Investment Management discusses with Yahoo …

Based on the publisher's description: Dan Niles of Niles Investment Management discusses with Yahoo …

Based on the publisher's description: Dan Niles of Niles Investment Management discusses with Yahoo Finance whether the AI stock sell-off resembles the dot-com crash or 1997. He covers investor rallies during the 2000-2002 downturn, companies shifting from token maxing to cost cutting, and how midterm ballot measures against data centers changed the AI trade.

Read brief

What are the latest AI investment signals?

Latest AI investment signals: 1 funding rounds, 3 market updates, and 1 M&A transactions.

Primary Market – Funding Rounds

CompanyAmountRoundInvestors
Databricksseveral billion dollarsUnknown

Secondary Market – Market Updates

M&A – Mergers & Acquisitions

AcquirerTargetDeal ValueDeal Type
AmazonDecart$7.5BAcquisition

What are practical AI tips this week?

1 practical AI tips curated from Reddit communities and expert blogs. OpenAI released new prompt templates for GPT-6.2 Astra and Sol aimed at fixing common model behavior...

Prompt Tips

OpenAI released new prompt templates for GPT-6.2 Astra and Sol aimed at fixing common model behaviors such as overly brief answers, unnecessary follow-up questions, and excessive testing when coding, according to The Decoder.

Use OpenAI's published prompt templates for GPT-6.2 Astra and Sol to address overly brief answers, unnecessary follow-up questions, and excessive testing when coding.

Read brief