the week in AI, briefly. then briefly again.
Artificial Intelligence — briefly, then briefly again · tbb.ceo
listen tothe daily 0:00 –:––
/daily ·21 NOV 2026 ·SATURDAY ·1 MIN READ ·7 STORIES

AI Brief — Saturday, 21 November 2026

AI hardware controls tighten, reasoning defenses thicken, and an agent-funding supercycle shows early jitters.

01 / The Day

SATURDAY 21 NOV 2026, ranked

07

AI-verified Navier-Stokes proof closes major gap

NYU and Anthropic researchers used AI-assisted theorem proving to solve a 40-year-old edge case in the Navier-Stokes equations, correcting a flaw in a prior solution.

  • AI-guided proof verified in Lean 4 formal language
  • Addresses turbulence behavior in 3D fluid dynamics
  • First peer-reviewed AI-assisted resolution of a Millennium Problem component
Why it mattersAI-assisted mathematics is now contributing to century-old open problems, not just benchmark tests.

Nasdaq 100 posts worst week since 2020 on AI fears

The Nasdaq 100 fell 5.6% Friday and 10.4% for the week — its worst since the March 2020 COVID crash — as investors reassessed AI capital expenditure cycles.

  • AI infrastructure spend fears triggered institutional rotation
  • Nvidia, Microsoft, Meta led the selloff
  • Bond markets saw flight-to-quality as equities fell
Why it mattersWall Street is pricing the possibility that AI ROI timelines extend longer than the current capex cycle.

Cognition raises $2B+ at $75B valuation

AI coding-agent startup Cognition is in talks to raise over $2 billion at a $75 billion valuation, up from $48 billion in September.

  • Valuation tripled in under six months
  • Round led by existing backers including Founders Fund
  • Valuation now exceeds Palantir market cap at IPO
Why it mattersAutonomous software-engineering agents command the highest AI multiples in this funding cycle.

Apple enters device-control AI agent race

Apple unveiled a framework for AI agents to control apps and device interfaces, adding authentication and sandboxing layers that restrict third-party agent access.

  • Apple approach sandboxes agents to approved app surfaces
  • Third-party agents blocked from low-level system calls
  • Directly competes with OpenAI Operator and Anthropic Computer Use
Why it mattersPlatform owners are racing to define who controls the agent layer on consumer devices.

Cursor drops Chinese AI models over geopolitical risk

Code editor Cursor has phased out Kimi K2.5 and other Chinese-developed AI models, citing geopolitical risk and enterprise customer compliance requirements.

  • Kimi K2.5 removed from Cursor model menu
  • Move follows enterprise security audits by Fortune 500 customers
  • Other Western AI tools under pressure to follow
Why it mattersThe enterprise AI stack is fragmenting along geopolitical lines faster than any standards body can track.

GPT-6.2 reasoning harder to distill: rate drops to 0.2%

Researchers found attempts to extract GPT-6.2 chain-of-thought reasoning for distillation dropped from 54% to 0.2% success, following deliberate obfuscation measures.

  • Reasoning trace extraction blocked by output filtering
  • Competing models relying on distillation face a wall
  • OpenAI moat now partly technical, not just data-scale
Why it mattersOpenAI is building anti-distillation into its frontier models, raising the cost of competition.

US export controls expand to consumer-grade SoCs

New US export control rules now cover a broader range of semiconductors, including consumer system-on-chip designs previously exempt from licensing requirements.

  • Controls extend to chips with certain AI inference capabilities
  • Consumer SoC vendors like MediaTek and Qualcomm affected
  • Supply chain groups warn of collateral damage to non-AI products
Why it mattersChip controls have outgrown their original scope and now touch everyday consumer electronics supply chains.
Be subscriber #028 weekly AI email interest list
Past days

The Daily Archive