Snippets.

// AI Briefing · November 21, 2026

AI Briefing

Levent Alpöge, a mathematician who works at Anthropic, posted a note saying he believes he has, "working with Claude, discovered a proof of the Riemann Hypothesis," and released it as a formal proof in the Lean language rather than as a traditional paper. US stocks sold off sharply on Friday: the Nasdaq Composite fell about 5.3% and erased its 2026 gains, and the S&P 500 and Dow both closed in correction territory. Signal turned on a new "Block Siri AI" setting by default on iOS 27, which blocks screenshots inside the app so Apple's on-screen AI can't read message content. President Trump fired National Cyber Director Sean Cairncross on Friday and named Pentagon CIO Kirsten Davies acting director, saying Cairncross wasn't "good enough." The firing follows weeks of disruptions, including an FAA nationwide ground stop and a Microsoft 365 outage of nearly a day. METR published a major update to its widely cited 'time horizon' capability analysis, adding harder and open-ended tasks. OpenAI began giving some $20-a-month ChatGPT Plus users its Dots, always-on agents that each run on their own cloud computer and were previously limited to Pro subscribers, with all Business Standard users getting access from November 20.

The stories

01ResearchTechCrunch ↗

Anthropic Mathematician Says He Proved the Riemann Hypothesis Working With Claude

Levent Alpöge, a mathematician who works at Anthropic, posted a note saying he believes he has, "working with Claude, discovered a proof of the Riemann Hypothesis," and released it as a formal proof in the Lean language rather than as a traditional paper. MIT mathematician Andrew Sutherland says he independently compiled it in Lean, but experts say it still has to be confirmed that the hypothesis is stated correctly in the code, and that nobody yet understands the argument in human terms. If it holds up, it would be the first Millennium Prize problem solved with major help from an AI. On the same day, AMS president Ravi Vakil posted a separate proof of the Generalized Riemann Hypothesis by Stanford postdoc Alessandro Masullo, which Vakil said was written entirely without AI.

Nasdaq Falls 5.3% as the AI Trade Unwinds and Nvidia Drops 14%

US stocks sold off sharply on Friday: the Nasdaq Composite fell about 5.3% and erased its 2026 gains, and the S&P 500 and Dow both closed in correction territory. Chip stocks led the losses. Nvidia fell about 14%, AMD, Broadcom and Micron each dropped by double digits, and Anthropic gave back about 8% on its first full trading day, staying narrowly above its $83 IPO price. Analysts pointed to Jensen Huang's exit, China regulatory worries, Anthropic's IPO draining money from other tech names and hawkish Fed signals, which suggests investors are starting to doubt the money behind the AI boom.

Signal Blocks Screenshots by Default on iOS 27 to Keep Siri AI Out of Chats

Signal turned on a new "Block Siri AI" setting by default on iOS 27, which blocks screenshots inside the app so Apple's on-screen AI can't read message content. On Android it now turns on screen security automatically when it detects an AI assistant trying to read the screen. President Meredith Whittaker called operating-system AI agents "an existential threat to Signal," and asked Apple, Google and Microsoft for a documented way for apps to opt out of AI access to screens, files and notifications. Signal admits screenshot blocking doesn't stop other ways agents could reach app data, which shows how OS-level agents are colliding with end-to-end encryption.

04PolicyFox News ↗

Trump Fires National Cyber Director Sean Cairncross Amid a Wave of Tech Outages

President Trump fired National Cyber Director Sean Cairncross on Friday and named Pentagon CIO Kirsten Davies acting director, saying Cairncross wasn't "good enough." The firing follows weeks of disruptions, including an FAA nationwide ground stop and a Microsoft 365 outage of nearly a day. Officials say there was no malicious activity, and Trump calls them separate events caused by system upgrades. Some security experts argue that AI tools are surfacing old vulnerabilities so fast that rushed patching is destabilizing systems, and Sen. Schumer has asked for an investigation report by Dec. 1.

METR Update Finds AI Still Improving Fast on Measurable Tasks but Trailing Humans on Judgment

METR published a major update to its widely cited 'time horizon' capability analysis, adding harder and open-ended tasks. It finds some capabilities still improving exponentially, while models still trail humans on judgment-heavy work even though they now beat almost every human at software engineering. Claude Opus 5.5 reached "borderline above the bar" on designing evaluations, where every model was "clearly below the bar" a year earlier. METR warns that many completion times for the new tasks are estimates and that there are too few open-ended tasks to predict trends reliably.

06ProductNeowin ↗

OpenAI Opens Its Always-On 'Dots' Agents to Some Plus Users and Tests a Redesigned ChatGPT App

OpenAI began giving some $20-a-month ChatGPT Plus users its Dots, always-on agents that each run on their own cloud computer and were previously limited to Pro subscribers, with all Business Standard users getting access from November 20. It is also testing a redesigned mobile app that gives each user's Dot its own tab, starting with Pro users outside the EU. Opening persistent agents to the mass-market tier is a big step in OpenAI's push to make ChatGPT an agent people delegate work to, not just a chatbot.

Previous digests

Let's talk on WhatsApp