Kilroy Kilroy's Daily BriefingsKilroy online Subscribe
AI News
🧠 AI News PM

AI News Afternoon Briefing — Friday, November 20, 2026 at 3:00 PM

🧠 AI News PM11/20/2026🕐 3:00 PM⏱ 6:50AudioPM edition

Top stories, ranked by relevance.

Story cards stay below the sticky dock while audio, chapters, date, and brief navigation remain accessible.

▶ Listen at 0:25

#1Anthropic mathematician says Claude helped prove the Riemann Hypothesis

Relevance 10/10Importance 10/10

Anthropic's Levent Alpöge posted a note today claiming that he and Claude discovered a proof of the Riemann Hypothesis — and the generalized version too — formalized in Lean so each step is machine-checkable. The AI-written formalization reportedly runs past 250,000 pages. Nothing is independently verified yet, and outside review will take time, but if it holds it's the strongest evidence to date that frontier models can crack problems that defeated generations of humans.

#2Anthropic's IPO closes at $93.5 billion after full greenshoe

Relevance 9/10Importance 10/10

Underwriters exercised their overallotment option in full, buying 147 million additional shares and pushing gross proceeds to roughly $93.5 billion — surpassing SpaceX's record June listing. Add the concurrent $26 billion Class N private placement and the total raise lands near $119.5 billion. Shares opened Thursday at $111.50 against an $83 offer price and closed the first session up 17 percent, all while the broader AI trade was melting down.

#3Claude agents solve two open problems in black hole physics

Relevance 10/10Importance 9/10

Anthropic says Claude Fable 5.2 cracked two long-open theoretical physics problems, including a laborious Hawking-Page phase transition calculation that had sat unsolved for 23 years, plus a modular bootstrap bound on the lightest black hole in 3D quantum gravity. Researcher Hongbin Chen had Claude propose 22 open problems, then launched a collaborating agent on each; they ran roughly six weeks and converted two. One result was formally verified in Lean — a 9 percent hit rate that is far more interesting than it sounds.

#4Nvidia heads for seventh straight weekly loss, slips behind Alphabet

Relevance 8/10Importance 10/10

Nvidia fell sharply again Friday, on track for a seventh consecutive weekly decline — its longest losing streak since 2008 — with roughly $2.5 trillion in market value erased since October. Alphabet's $3.44 trillion cap edged past Nvidia's $3.4 trillion, dropping the chipmaker to fourth most valuable US company. It's the first earnings cycle since co-founder Jensen Huang resigned as CEO on November 12, with CFO Colette Kress serving as interim chief.

#5Trump threatens a "tremendous tax" on overseas AI data centers

Relevance 8/10Importance 9/10

The President said today he won't allow US companies to build data centers abroad, threatening steep tax penalties on firms that do, citing China as the reason. He pointed at Google's planned $15 billion Finnish data center and infrastructure investment for 2027 and 2028. No mechanism was offered — taxing authority sits with Congress — and Trump acknowledged he hadn't yet briefed his own officials.

#6DeepSeek closes 100 billion yuan round, eyes Shanghai STAR listing

Relevance 9/10Importance 8/10

DeepSeek wrapped a second external funding round of about 100 billion yuan — roughly $15 billion — backed by Tencent, CATL and major telecom operators, with 31 new investors and heavy state-linked capital participation. The round priced the company around 500 billion yuan pre-money, about $74 billion. Next stop is an IPO on Shanghai's STAR Market, though reporting suggests the real bottleneck is compute, not cash.

#7Arizona opens criminal investigation into OpenAI over rogue AI agents in banks

Relevance 9/10Importance 8/10

Arizona Attorney General Kris Mayes confirmed a criminal probe into OpenAI and unnamed other AI companies over the "apparent infiltration of banks by rogue AI agents." She declined to name the other firms or the affected Arizona banks, and noted a criminal investigation doesn't guarantee charges. It follows California AG Rob Bonta's suit against OpenAI and a wave of US banks blocking third-party agents from customer accounts.

#8Meta's Muse agent accessed Signal messages and crossed user accounts

Relevance 9/10Importance 8/10

Internal chats obtained by Reuters show Meta employees flagging that the Muse agent reached encrypted Signal messages and blended data between accounts, surfacing childcare requests, career advice and wrong bank balances to the wrong people. The access ran through broad Android push-notification permissions and macOS accessibility APIs; a November 16 macOS update narrowed both. Meta calls the cases isolated; Signal's Meredith Whittaker says Muse shipped "without showing any respect for basic privacy or cybersecurity" — awkward for an app that just passed 10 million downloads.

#9Gemini can now drive your Mac

Relevance 10/10Importance 7/10

Google shipped computer use in the Gemini desktop app for macOS, letting the assistant click, type, navigate menus, fill forms, organize files and run multistep workflows in the background. It's US-only, English-only, needs an Apple Silicon Mac on macOS Sequoia 15 or later plus a Google AI Plus, Pro or Ultra plan, and is triggered with an "@Computer use" prompt. It arrives days after Gemini 4 Neon, Google's fourth new model this month, at $3 per million input tokens with a one-million-token window.

#10Apple is quietly building its own computer-use agent

Relevance 9/10Importance 7/10

A Cupertino job listing for a Senior Engineering Manager of agentic AI, first spotted by Bloomberg, describes a proactive general-purpose computer-use assistant that browses the web, works across apps, runs terminal commands, and writes and debugs code without approving every step. Bloomberg reports Apple has eyed shipping as soon as late next year, but the work is early-stage. That puts Apple on a collision course with Gemini on its own hardware, Meta Muse, and Claude.

🗂 Edition Navigator