Daily Summary of AI · Final editionFriday 20 November 2026 · UTC
AI News: 20 November 2026
Riemann claim and AI selloff hit Nvidia as Trump threatens data center tax
Anthropic mathematician Levent Alpöge says Claude helped prove the Riemann hypothesis, with a Lean proof that some mathematicians say compiles but still needs verification. AI investment worries renewed a selloff, with Nvidia falling for a seventh week and Anthropic shares slipping near their IPO price. Trump threatened a tax on US firms building AI data centers abroad, while Meta re-evaluated its 2030 net-zero goal after emissions rose.
Meta re-evaluates 2030 net-zero goal as AI emissions surge
Meta is re-evaluating its 2030 net-zero emissions goal after its 2026 sustainability report showed total emissions rose from 15.6 million metric tonnes in 2024 to more than 21.7 million in 2025. Executives said AI infrastructure demand grew faster than projected. The increase, nearly 40 percent, is the highest reported by any of the four major US tech giants.
Nvidia falls for seventh week as Huang's exit and AI worries weigh
Nvidia shares fell on Friday, with the drop reported at 12% or 14%, extending a seventh straight day and a seventh consecutive weekly decline, the longest since 2008. The stock is down 46% over the period, with its market cap down by about $2.6 trillion and now at roughly $3 trillion. Investors are weighing the company's direction after co-founder Jensen Huang resigned and reports that the SEC is exploring potential accounting issues.
Huang to testify publicly before Senate banking panel
Former Nvidia CEO Jensen Huang will testify publicly before the Senate banking panel, a spokesperson for Chairman Tim Scott said Friday. Scott confirmed earlier this week that Huang agreed to appear, but the date has not been announced. Huang, who resigned over extramarital affairs including one with a suspected Chinese spy, also agreed to a closed-door, transcribed interview with three House committees on Dec. 1.
Trump threatens 'tremendous tax' on US firms building AI data centers abroad
Trump said the US will ban firms from building AI data centers abroad and impose a 'tremendous tax' on those that do, a proposal he first announced publicly at a White House press conference. He gave few details and said his staff had not been told, even as states and cities restrict new data centers and tech firms already operate overseas. The threat aims to keep AI data centers in the US, but a new Democratic Congress could complicate it.
Anthropic mathematician claims Claude helped solve Riemann hypothesis
Anthropic mathematician Levent Alpöge says he has proved the Riemann hypothesis with Claude, a claim that would be a seismic AI achievement in mathematics. The proof, written in Lean, is on his website and has not been independently verified or accepted by the Clay Mathematics Institute. Alpöge says he told Anthropic CEO Dario Amodei late Sunday night and appears to have waited until after the company's IPO.
Bloomberg: AI-exposed job listings in China fell 10 percentage points faster
A Bloomberg analysis of nearly 10 million online job postings from July 2021 to June 2025 found that listings for jobs highly exposed to AI fell about 10 percentage points more than listings for less affected roles. The shift is adding pressure to China's already weak labor market, with entry-level workers and the creative industry reported to be feeling it.
Nvidia nears $500 million deal for Inception's diffusion models
Nvidia is nearing a deal to acquire Inception for at least $500 million, with performance bonuses for the startup's more than 30 employees. Inception builds language models using diffusion, refining full responses step by step instead of generating tokens one at a time, and recently released Mercury 3. The deal is meant to strengthen Nvidia's open source Nemotron models and physical models for robots.
NVIDIA releases Nemotron-Labs-Sigma, a continuous diffusion model family
NVIDIA has released Nemotron-Labs-Sigma, a family of 14 open checkpoints at 3B and 8B scale that denoises text in a 16-dimensional embedding space. The team calls it the first large-scale family of continuous diffusion language models, saying its smooth latent space makes few-step distillation and trajectory steering practical. NVIDIA's own evaluations show Sigma-8B averaging 71.16 on a six-task base suite, with a 32-step sampler at 70.18.
Cursor considers phasing out Chinese AI models over geopolitical risk
SpaceX's coding platform Cursor is considering phasing out Chinese AI models because of geopolitical risk, according to a report by House committees. Cursor, acquired by SpaceX at an implied $60 billion equity value in August, built Composer 2 on Moonshot AI's Kimi K2.5. The report says Chinese models are adopted for cost, customization and near frontier performance, and recommends a national American open weight initiative.
Anthropic raises $93.5 billion total as underwriters exercise overallotment
Anthropic's underwriters exercised their overallotment, bringing the company's IPO total to $93.5 billion and surpassing SpaceX's June record. The company sold 1,127,000,000 Class A shares at $83, including the additional 147 million shares available under the overallotment. Anthropic also closed a $26 billion private placement with AMD, Google, Microsoft, Nvidia and others, making gross proceeds $119.5 billion before expenses.
Signal blocks screenshots to stop Gemini and Siri from reading chats
Signal is restricting screenshots on Android and iOS to prevent AI agents from reading messages and other data. President Meredith Whittaker said agentic AI built into operating systems undermines Signal's ability to independently guarantee privacy. Users can restore screenshots in app settings, but that may expose content to Siri, Gemini and other AI assistants. Signal calls the fix incomplete and asks OS makers for tools and documentation.
Meta's AI assistant Muse is being used by Instagram users to identify and unfollow people who do not follow them back, creating a trend of automated social-media cleanses. One user with more than 7,600 followers asked Muse to list non-followers and unfollow 151 accounts, receiving an Excel sheet with names and profile links. The feature has spread across X, Instagram and Facebook, with users sharing prompts and celebrating unfollow counts.
Anthropic shares fall near IPO price as AI sell-off hits global markets
Anthropic shares fell on their second day of trading, dropping as much as 12% at one point to about $85, near the $83 IPO price. The decline helped push the Nasdaq Composite to erase its 2026 gains, while Asian markets recorded their worst week since 2020. Investors are questioning whether the huge spending on AI infrastructure will pay off.
OpenAI tightens defenses against adversarial distillation of GPT-6.2 models
OpenAI says it has tightened defenses against adversarial distillation, in which a model's outputs are used to train another model without comparable safeguards. In internal tests, the average success rate for extracting protected reasoning from earlier prompts fell from 54.0 percent on GPT-6 Astra to 0.2 percent on GPT-6.2 Astra. Attacks on active requests fell from 22.9 to 0.9 percent, and OpenAI is bringing selected changes to GPT-6 Astra.
Anthropic shares fall in premarket as AI sell-off hits chipmakers
Anthropic shares fell about 8% in premarket Friday after a record debut, as AI investment worries renewed a sell-off. Nvidia and AMD shares also dropped, with AMD down more than 7% after a short seller alleged its firmware appeared in U.S.-blacklisted companies. The House Foreign Affairs Committee released documents showing a Commerce Department subagency approved sales of more than 879,000 AI chips to Chinese companies.
16-year-old's AI missile startup joins Y Combinator
Latvian-born Henrijs Dandzbergs, 16, was admitted to Y Combinator's fall 2026 cohort for Polemos Industries, an AI-guided missile startup. The company is developing precision-guided missiles that can be cheaply made, fired from portable launchers, and resist jamming, with an ambition to lower the cost per kill. Dandzbergs is likely the youngest solo founder in Y Combinator's history, and the startup won half a million dollars.
AI coding tools speed work, but some developers miss the struggle
AI tools are now part of daily work for most software developers, with 83% of respondents in a 2026 Stack Overflow survey saying they use at least one. Some engineers say the tools speed up writing, debugging and testing, but reduce the satisfaction of solving hard problems by hand. Others note the tools lower barriers to entry, while still warning that core skills can be lost if developers rely too much on prompts.
SoftBank Group shares plunged 16.16% on Friday, leading a broader sell-off in Asian technology stocks after Anthropic's market debut. The decline highlighted investor concerns over SoftBank's $64.6 billion investment in OpenAI and whether large AI infrastructure bets will generate returns. Japanese semiconductor stocks, South Korean Nvidia suppliers and other Asian tech shares also fell, after the Nasdaq Composite dropped more than 2% overnight.
Apple appears to be building an AI agent that can use a Mac
Apple appears to be developing its own AI agent that can browse the web, work across apps, run commands and write code. A job listing seeks a senior engineering manager to lead an agent that can use a computer like a capable person. Bloomberg says the service could arrive as early as late next year, while Apple is making it harder for third party agents to control Macs, with macOS 27.3 requiring 14 steps instead of four for full access.
Anthropic founders cash out nearly $2B in record IPO
Anthropic became a public company on Thursday, raising $81.3 billion in a record-breaking IPO. Its seven founders sold about 3.4 million shares each at the upgraded price of $83, taking in about $280 million each before underwriting fees and taxes. The company said each founder has pledged 80% of their personal equity to philanthropy, though the commitments are voluntary and revocable.