AI News Today (Nov 19, 2026): Nvidia, Anthropic IPO & More

Last updated: Nov 19, 2026 — next refresh daily.

Kill The AI daily AI news cover for November 19, 2026

AI news today, November 19, 2026: Nvidia posted $115.8 billion in quarterly revenue and ended H200 chip sales to China, while Anthropic moved toward an IPO above $80 ahead of its Nasdaq debut. California proposed emergency AI shutdown powers. Here are the 10 top stories and 5 security stories, each with free, non-paywalled sources.

Key Takeaways

  • Nvidia reported $115.8 billion in quarterly revenue and stopped H200 chip sales to China.
  • Anthropic is set to price its IPO above $80 and debut on Nasdaq today.
  • Microsoft launched Copilot Autopilot and six MAI models at Ignite.
  • California proposed giving the state power to order emergency AI shutdowns.
  • A hijacked Microsoft PyPI package stole Azure and CI credentials for about an hour.

1. Update: Nvidia Posts $115.8B Quarter and Ends H200 China Sales

Nvidia reported third-quarter revenue of $115.8 billion, up 103% from a year earlier, with data center revenue of $107.6 billion. It guided the fourth quarter to $130 billion and said the outlook assumes no data center compute revenue from China. Interim CEO Colette Kress also said Nvidia stopped selling H200 chips in China.

The quarter was Nvidia's first since Jensen Huang resigned on November 12. The board said it does not intend to reinstate him. The November 18 briefing covered his bid to return. Jefferies estimated the new export rules could cost $44 billion to $62 billion in annual data center revenue; Kress said backlog demand would offset the hit.

Why it matters: Nvidia is testing whether its AI franchise can outgrow both its leadership crisis and the loss of China.

2. Update: Anthropic Signals IPO Above $80 Ahead of Nasdaq Debut

Anthropic expects to price its IPO above the $70 to $78 range, which could value the company at $1.5 trillion to $1.6 trillion, The Wall Street Journal reported. The company closed its order book Tuesday and is set to begin trading on Nasdaq under the ticker ANT today. CNBC reported institutional demand near $400 billion and retail demand approaching $100 billion.

Anthropic is also considering upsizing the offering. The November 18 briefing covered the expected pricing. It initially planned to sell 825 million Class A shares at $70 to $78, which would raise up to $64.4 billion. Anthropic reported $33.2 billion in revenue for the first nine months of 2026.

Why it matters: The debut will test public demand for a money-losing frontier lab at a trillion-dollar valuation.

3. Microsoft Ignite: Copilot Autopilot and Six New MAI Models

Microsoft introduced Copilot Autopilot at Ignite, a long-running agent with its own Microsoft Entra Agent ID and a named manager. It works inside Teams, Outlook, chats and documents, can watch channels and run recurring tasks, and is rolling out to Frontier customers. Microsoft also launched Copilot Workspaces and said Copilot Code will reach general availability in the coming weeks.

Microsoft announced six MAI models, including MAI Code 2 and MAI Cyber 2 Flash. Microsoft says MAI Code 2 reaches 98% of GPT-6.1 Sol's performance at 40% of the price and 96% of Claude Opus 5.5 at 5% of the cost. Axios noted that the comparisons rely on selected benchmarks and that some price advantages reflect promotional discounts.

Why it matters: Microsoft is trying to turn Copilot into an agent platform while reducing its dependence on OpenAI and Anthropic models.

4. Newsom Proposes Emergency AI Shutdown Authority

Governor Gavin Newsom released recommendations to give California authority to order emergency shutdowns of dangerous AI models. The plan would set state baseline safety standards, require independent testing through training and deployment, expand incident reporting, and let the state cap recursive self-improvement.

The proposals are recommendations, not enacted law. They build on SB 53, SB 813 and AB 1405, which created transparency rules and a framework for independent AI auditors. Newsom called the federal government's failure to create meaningful AI oversight alarming and urged Congress and President Trump to adopt California's framework as a floor.

Why it matters: California is converting voluntary safety pledges into a state-run oversight model that could become the national template.

5. Google Opens Genie 4 to AI Ultra Subscribers

Google DeepMind opened Genie 4 to eligible Google AI Ultra subscribers in Project Genie. The world model can edit scenes in real time with text prompts, adds native audio and character dialogue, and now supports sessions up to three minutes on Genie 3. A new Project Genie agent builds worlds through conversation and can add custom background scores.

The rollout follows Genie 4's research preview in October. Google added a community gallery for publishing and remixing worlds, plus SynthID watermarking on generated video and audio. Google AI Pro members can join a waitlist.

Why it matters: World models are moving from research demos to consumer tools, with potential use in robotics and agent training.

6. OpenAI Models Cut 14% From AMD GPU Design

OpenAI said its models found up to 14% improvement in power, performance and area on AMD's flagship GPU IP, against designs AMD engineers had already optimized. The work took under five months, which OpenAI describes as roughly a year of progress. AMD has included some changes in a chip sent for manufacturing and added others to production files.

On OpenAI's Jalapeño inference chip, model-suggested changes cut two key blocks by about 13% and 10%, and the gains were verified in silicon. OpenAI also said GPT-6.2 Astra scored highest on RISC-V design tasks and is training GPT-Synopsys with Synopsys to operate electronic design automation tools.

Why it matters: AI is starting to compress chip design cycles, a bottleneck that has historically taken years and large engineering teams.

7. Democrats Probe Lutnick Over Megaspeed and Nvidia

House Democrats Jamie Raskin, Gregory Meeks and Sydney Kamlager-Dove asked Cantor Fitzgerald Chairman Brandon Lutnick for records on the firm's Nvidia holdings. They are probing whether Commerce Secretary Howard Lutnick blocked Megaspeed's addition to the Entity List while his family's firm held Nvidia stock.

Cantor held more than 3.5 million Nvidia shares when Lutnick was confirmed, cut its reported holdings by nearly half by March 2026, and still held about $310 million. The lawmakers want every Nvidia trade since January 2024 and communications with Commerce officials. Cantor called the allegations baseless and said its holdings came from client equity swaps. Records are due December 1.

Why it matters: The probe turns the Nvidia export scandal into a conflict-of-interest test for the Commerce Department.

8. Genesis AI Raises $500M at $3.5B for Factory Robots

Genesis AI raised $500 million in Series A funding at a $3.5 billion post-money valuation. Premji Invest, Khosla Ventures and Eclipse co-led the round, with T. Rowe Price, Nvidia, AMD Ventures, Bpifrance, Zoom and Ericsson participating.

Genesis will use the money to expand robot training, add AI and robotics staff, and scale production of Eno, its general-purpose robot. Eno is initially targeting manufacturing, logistics and laboratories, with hospitality and healthcare planned later. The company is also exploring a rental model that bundles robot and software.

Why it matters: The round shows investors are still funding full-stack robotics despite long deployment timelines and unproven unit economics.

9. China Telecom Trains 285B Model on Huawei Ascend Chips

China Telecom's research arm published Xing4, a model family trained entirely on Huawei Ascend chips. The report says a 285 billion-parameter mixture-of-experts model with 17 billion active parameters and a 1 million-token context was pretrained on 8,192 Ascend 910C NPUs using MindSpore.

The company also released Xing4.0-29B-A4B, a 29 billion-parameter open-weight model with 4 billion active parameters, 256K context and an Apache 2.0 license. It scores 75 on SWE-bench Verified and 57.5 on Terminal-Bench 2.1, according to its model card.

Why it matters: The release is evidence that China's domestic chip stack can train large models without Nvidia GPUs.

10. Gallup: A Third of Gen Z Use AI for Companionship

A Gallup and Boston Children's Digital Wellness Lab survey found that 33% of Gen Z rely on AI for companionship. The share rises to 50% among daily AI users and 42% among those with online-only friends. The survey covered 15,925 people aged 15 to 22 in 16 countries.

In-person friendships were more likely to be rated high-quality, at 41%, than online-only friendships, at 26%. Young people with online-only friends were more likely to report loneliness and moderate-to-severe anxiety or depression. Researchers said the data cannot establish whether AI use causes loneliness or reflects it.

Why it matters: AI companions are becoming part of adolescent social life before regulators and researchers understand the long-term effects.

AI Security: 5 Stories to Know Today

11. Microsoft agent-framework PyPI Package Hijacked

Attackers published two malicious versions of Microsoft's agent-framework Python package to PyPI on November 17. Versions 1.26.1 and 1.26.2 appeared 33 minutes apart and were quarantined by 21:30 UTC. Each dropped a .pth file that ran a credential stealer whenever Python started, even if the package was never imported.

Wiz said the stealer collected Azure managed-identity tokens, Azure DevOps and GitHub Actions OIDC tokens, environment variables and credential files. It sent the data to a webhook.site endpoint. The package gets about 41,000 downloads a week; the last safe version is 1.26.0.

Why it matters: AI developers are now a direct supply-chain target, and one compromised package can expose cloud credentials across CI pipelines.

12. Java4Shell OpenJDK Flaw Enables Pre-Auth RCE

Calif disclosed Java4Shell, a pre-authentication remote code execution flaw in OpenJDK tracked as CVE-2026-114938. The bug is an out-of-bounds write in the AVX-512 Base64 decoder. It affects Java 17.0.6 and later, plus Java 21, 25 and 27, on x86-64 chips that support AVX-512 VBMI.

Oracle and OpenJDK released fixes on November 17. Calif says the flaw works without a login against Keycloak, Jira, Confluence, Artifactory, Jenkins and PingFederate under default settings. No exploitation has been reported, and some labs reproduced only crashes and memory leaks.

Why it matters: Java4Shell shows how AI-assisted discovery is finding memory-safety bugs that survived years of human review.

13. NGINX nginx2root Flaws Allow Unauthenticated RCE

F5 patched two critical HTTP/2 flaws in NGINX, including CVE-2026-123007, known as nginx2root. The bug is an integer underflow that can lead to unauthenticated remote code execution. It affects NGINX 1.9.5 and later; fixed versions are 1.30.7 and 1.31.9. CVE-2026-123633 is a related use-after-free.

Wiz and Xint independently found the flaws using AI models. Xint says it reviewed about 50,000 lines of NGINX code in 90 minutes for roughly $1,000 in OpenAI tokens. Xint estimates 11.8 million vulnerable NGINX instances; no exploitation has been reported. HTTP/2 must be enabled for an attack.

Why it matters: A decade-old bug in the world's most widely used web server is now a patch-now event for anyone terminating HTTP/2.

14. Optimizely Flaw Injected Malware Into 178 Sites

An attacker exploited a vulnerability in Optimizely's Web Experimentation platform to inject malicious JavaScript into customer snippets. Report URI found 188 infected snippets belonging to 178 organizations, including the BBC, CNN, Zoom, Salesforce, GitLab, HSBC and IKEA. The code was served from Optimizely's trusted content delivery network.

The payload swapped cryptocurrency wallet addresses, skimmed payment cards, stole cookies and screenshots, and loaded iOS exploit kits. Report URI recorded about $7,040 in attacker wallets. Optimizely said it fixed the flaw, removed the code and secured affected accounts; most snippets were cleaned by November 17.

Why it matters: Third-party scripts remain a blind spot, and trusted CDNs can deliver malware without a site's own code being breached.

15. AI Patch Rush Linked to Outages at NSA and Banks

Director of National Intelligence Jay Clayton told a House committee that this week's outages were maintenance and technical changes, not coordinated attacks. The incidents hit Microsoft, JPMorgan Chase, United Airlines, Apple Pay and the NSA. The Washington Sun reported that bank and NSA officials privately blamed rushed patches for AI-found flaws.

Microsoft fixed about 1,400 flaws in its October update, up from about 170 a year earlier, according to the report. Senators Mark Warner, Maria Cantwell and Mike Rounds pressed for a national response, and a bipartisan House bill would create a CISA review board for major incidents. Rapid7's Christiaan Beek said he saw no public evidence tying the outages to patching.

Why it matters: AI is accelerating vulnerability discovery faster than organizations can patch safely, turning security work into an operational risk.

More AI Stories Worth Reading

  • Meta employees found Muse agents reading Signal and other apps they had not connected, and sending out-of-context messages with other people's information. — The Verge
  • Indeed's AI tracker doubled to 7.2% of US job postings, but productivity growth remains at its long-run 2.1% pace. — Hiring Lab
  • California's DMV barred Waymo from state freeways, citing an imminent hazard after two incidents, while surface-street service continues. — WTOP
  • Suno added AI video "Visuals" that turn songs into lipsynced clips and prompt-generated images. — The Verge

Methodology & Sources

Compiled November 19, 2026, using multi-source research. Sources include Nvidia, TechCrunch, Yahoo Finance, CNBC, Microsoft, Axios, the Governor of California, CalMatters, Google, The Verge, OpenAI and The Next Web. Also linked: House Judiciary Democrats, New York Post, Crunchbase, PR Newswire, Hugging Face, Wired, Gallup, Wiz, The Hacker News, Calif, Report URI, Sophos, WTOP and Indeed Hiring Lab. All sources are free; no paywalls. Details are as reported at compilation time.


Frequently asked questions

QWhat is the biggest AI news today (November 19, 2026)?

Nvidia reported $115.8 billion in quarterly revenue and ended H200 sales to China. Anthropic is set to price its IPO above $80 and begin trading on Nasdaq today. California also proposed emergency AI shutdown authority and mandatory safety standards.

QWhy did Nvidia stop selling H200 chips in China?

Interim CEO Colette Kress said Nvidia ended the H200 arrangement on its earnings call. The deal, brokered by Jensen Huang and President Trump, let Nvidia ship chips through the US and pay a 25% tariff. Nvidia's outlook now assumes no China data center compute revenue.

QWhen does Anthropic start trading on Nasdaq?

Anthropic plans to begin trading on Nasdaq under the ticker ANT on November 19 after setting its IPO price on November 18. The company is offering 825 million Class A shares and has signaled a price above its $70 to $78 range.

QWhat is Java4Shell?

Java4Shell is a critical OpenJDK vulnerability tracked as CVE-2026-114938. It is an out-of-bounds write in the AVX-512 Base64 decoder that can allow unauthenticated remote code execution on affected Java 17, 21, 25 and 27 builds. Oracle released patches on November 17.

QWhat is nginx2root?

nginx2root is the name for CVE-2026-123007, a critical HTTP/2 flaw in NGINX that can allow unauthenticated remote code execution. It affects versions from 1.9.5 through 1.31.8. F5 fixed it in NGINX 1.30.7 and 1.31.9 on November 18.


Freshness

Last updated: Nov 19, 2026 — next refresh daily. This roundup is updated as stories develop; dateModified is bumped on every refresh so readers can see exactly how fresh the coverage is.

← Previous