AI News Today (Nov 20, 2026): Anthropic, Gemini 4 Neon & More
Last updated: Nov 20, 2026 — next refresh daily.
AI news today, November 20, 2026: Anthropic raised more than $81 billion and opened at $111.50, while Google released Gemini 4 Neon and put computer use inside the Mac. Four Maryland students sued over an AI gun alert. Here are the 10 top stories and 5 security stories, each with free, non-paywalled sources.
Key Takeaways
- Anthropic raised more than $81 billion in a record IPO and opened at $111.50, valuing it above $2 trillion.
- Google released Gemini 4 Neon with a 1 million-token context and $3/$15 per million token pricing.
- Gemini can now control Mac apps and files for U.S. subscribers, with user approvals.
- Four Maryland students sued after an AI alert mistook a Doritos bag for a gun.
- The European Commission confirmed an ARTEX AI-assisted attack that took HR systems offline for over 10 days.
1. Update: Anthropic Opens at $111.50 After Record $81B IPO
Anthropic raised more than $81 billion in a record IPO and opened on Nasdaq at $111.50 a share, 33% above its $83 offer price. The opening valued the Claude maker above $2 trillion, making it the sixth-most-valuable U.S. company, ahead of Meta, Tesla and Walmart. Anthropic sold more shares than planned after demand outstripped the offering, the BBC reported.
SpaceX raised $75 billion in June, making Anthropic's offering the largest on record. Anthropic expects about $220 billion in cash after the listing and concurrent private placements, TechCrunch reported. The company also reported $33 billion in revenue and $104 billion in net losses for the nine months to September, according to the BBC. The November 19 briefing covered pricing.
Why it matters: The debut tests whether public markets will fund a loss-making frontier lab at a scale once reserved for mature technology giants.
- Source: Anthropic hits Nasdaq exchanges at $111, 33% above IPO price — TechCrunch
- Source: Anthropic raises $81bn in record-breaking stock market debut — BBC
2. Google Releases Gemini 4 Neon With 1M Context and Lower Prices
Google DeepMind released Gemini 4 Neon, its fourth model in seven weeks, with a 1 million-token context window and support for text, images and video. Neon scores 61 points at its top setting on the Artificial Analysis Intelligence Index, third behind OpenAI's GPT-6.2 Astra and Anthropic's Claude Fable 5.2. Google says it produces more secure code and resists prompt injection.
Google priced Neon at $3 per million input tokens and $15 per million output tokens. That is 25% below Carbon's list price but 50% above its promotional price through November 30. The Decoder reports Neon runs at about 110 tokens per second, nearly double Carbon's 57, and costs $6.44 per task at its highest setting versus $7.76 for Carbon.
Why it matters: Google is compressing release cycles and cutting list prices to defend the mid-tier, where most agent workloads actually run.
- Source: Introducing Gemini 4 Neon — Google DeepMind
- Source: Google's latest AI model is faster, cheaper and smarter than its flagship predecessor from just weeks ago — The Decoder
3. Gemini on Mac Can Now Control Apps and Files
Google added computer use to the Gemini app for macOS, letting the agent work across local files, desktop apps and the web. The feature is available in English for U.S. users on AI Plus, Pro and Ultra plans, and shows a live preview while the user's screen and cursor stay free. It works while the user does something else.
Users can pause, stop or approve actions before they run, Google says. The launch lands ahead of Apple's macOS 27.3 beta restrictions on full disk access for apps that use AI agents. OpenAI, Anthropic and Meta already ship similar agentic desktop features. Google did not say when it would reach other markets or platforms.
Why it matters: Agentic AI is moving from chat windows into direct control of the desktop, just as platform owners tighten the permissions it needs.
- Source: Gemini on Mac can now use your computer — The Verge
- Source: Google gives Gemini the ability to use your Mac for you — TechCrunch
4. Doritos Bag Mistaken for Gun Leads to Lawsuit Against AI Vendor
Four current and former Black students at Kenwood High School in Baltimore County sued Omnilert, the school system and police. An AI gun-detection system had flagged a bag of Doritos as a firearm with 90% confidence. Armed officers ordered the students to their knees, handcuffed and searched them in October 2025. They were waiting for rides.
The lawsuit, filed November 17 in Baltimore County Circuit Court, alleges unlawful detention, constitutional violations and negligence. Maryland's inspector general found more than 99% of the district's AI alerts were false positives. The alert continued after staff and police reviewed the image and saw no gun. Omnilert said it was reviewing the lawsuit and was "truly sorry."
Why it matters: The case tests whether schools and police can treat an AI alert as probable cause before a human verifies it.
- Source: Lawyers' Committee Files Lawsuit on Behalf of Black Students Detained and Searched Due to AI Alert Falsely Identifying a Bag of Chips as a Firearm — Lawyers' Committee for Civil Rights Under Law
- Source: Four Maryland high school students sue after AI mistakes Doritos bag for gun — The Guardian
5. Nadella Says 60,000 Buyouts Turn Every Job Into Agent Management
Microsoft CEO Satya Nadella said the company offered buyouts to 60,000 employees in October so workers could choose whether to join its AI-first reorganization. He told the DealBook Summit that Microsoft is not replacing engineers, but expects every job to become a form of managing AI agents. He said the buyout gave employees agency in making choices.
Nadella pointed to Microsoft's Cowork team, where 11 developers use AI to build a software factory that produces the product. He also said AI finds software bugs faster, but the resulting patch velocity can destabilize old infrastructure. He called the recent Outlook outage a "perfect storm." Microsoft's stock has fallen nearly 9% since its November 4 record close.
Why it matters: Microsoft is defining the AI-era job as agent management, a template other large employers are likely to follow.
- Source: Microsoft CEO Satya Nadella explains why he asked 60K employees to quit — TechCrunch
- Source: Microsoft CEO Satya Nadella calls Outlook outage a 'perfect storm.' Here's what went wrong — CNBC
6. Update: China's AI Firms Hit as Chip and Cloud Access Stop
New U.S. export rules now require licenses for advanced AI chips used in overseas data centers serving Chinese companies, including Singapore, Malaysia and Indonesia. Moonshot AI canceled a $150 million deal for Nvidia B300 capacity in Indonesia, while TSMC and Samsung halted some shipments to Chinese customers. Operators get a 90-day transition and must screen customers for Chinese ties.
Tencent President Martin Lau said the rule "sets everyone back in China," and Alibaba warned it could affect some global data center expansion plans. A study cited by ABC News estimates remote access could account for about one-third of China's AI compute by the end of 2027. Beijing called it economic coercion. The November 17 briefing covered the rule change.
Why it matters: Washington is now policing compute access, not just chip sales, which raises costs for Chinese labs and their overseas cloud partners.
- Source: Chinese tech firms face a double hit as chip shipments stop and overseas cloud access disappears — The Decoder
- Source: Chinese tech firms fear AI hits from US curbs after Nvidia fallout — ABC News
7. IEA: Data Center Power Use Could Double by 2030
The International Energy Agency (IEA) said data centers could consume about 1,000 terawatt-hours of electricity a year by 2030, roughly double today's level. AI drives three-quarters of the growth. The agency said grids need 26 million kilometers of new lines by 2035, plus refurbishment of 16 million kilometers. The forecast draws on satellite imagery of 93 large AI factories.
The IEA expects data-center emissions to rise 70% by 2030. The U.S. is projected to get about 60% of data-center power from gas, while China would get about half from coal. Grid spending could rise 75% over the next decade, and connection queues already exceed peak demand in some regions. Copper, cable and transformer supplies are already tight.
Why it matters: Power and grid capacity, not chip supply, are becoming the binding constraint on how fast AI can expand.
- Source: Data center electricity use could double by 2030, IEA says — CNBC
- Source: Datacentres to create 70% more carbon emissions by 2030, IEA report warns — The Guardian
8. Palo Alto Buys Applied Compute for $4.5B as Earnings Beat
Palo Alto Networks beat fiscal first-quarter estimates with $3.44 billion in revenue, up 39% from a year earlier, and agreed to buy Applied Compute for $4.5 billion in stock. Shares rose about 6% as the cybersecurity company pushed beyond firewalls into training and serving open-weight AI models. The quarter ended October 31, and Palo Alto raised its full-year guidance.
Applied Compute, founded in 2025 by former OpenAI researchers, raised $350 million at a $3.25 billion valuation this month and says its annual recurring revenue has topped $100 million. The deal is expected to close in Palo Alto's fiscal third quarter; CEO Yash Patil will become co-chief AI officer. Applied Compute's platform handles post-training, inference and model routing.
Why it matters: Security vendors are buying their way into model training so they can control how enterprises deploy open-weight AI.
- Source: Palo Alto stock jumps on earnings beat and company's big bet on securing open-weight models — CNBC
- Source: Palo Alto Networks bets $4.5B on AI startup Applied Compute as earnings top forecasts — SiliconANGLE
9. Update: IMF Says AI Demand Will Push Interest Rates Higher
IMF Managing Director Kristalina Georgieva said AI demand and the Middle East energy shock will push global inflation to 4.7% this year and require tighter monetary policy in many economies. She told the APEC summit in Shenzhen that AI hardware now accounts for more than one-tenth of world goods trade. Countries need digital infrastructure and AI-ready workers, she said.
The European Central Bank, U.S. Federal Reserve and Bank of England have all raised rates this year. Georgieva said the energy shock will fade but "AI is for ever," making it a lasting driver of trade, productivity and growth. Markets expect further increases next year. The November 17 briefing covered her earlier warning on $1 trillion in circular AI financing.
Why it matters: AI is becoming a macroeconomic force that shapes inflation and borrowing costs, not just a technology trade.
- Source: AI demand will push up interest rates around the world, IMF head warns — The Guardian
- Source: IMF's Georgieva: Both energy and AI demand shocks are fuelling global inflation — FXStreet
10. Dutch Watchdog Tells Consumers Not to Buy AI Camera Glasses
The Netherlands' data protection authority, the Autoriteit Persoonsgegevens (AP), advised consumers not to buy AI camera glasses after testing six models priced from under €100 to over €500. Testers found glasses that claimed an ambulance was coming during a CPR simulation, misread medicine instructions and gave dating advice to an adult interested in a minor.
The AP also found security flaws that let someone nearby connect without permission, view recordings or operate the camera, and said recording lights could be covered with tape. Some glasses sent data outside the EU, the BBC found. It cannot ban the products itself; Dutch product safety authorities would have to act. Meta said it was considering a response.
Why it matters: Camera glasses are reaching consumers faster than tests, privacy rules and security standards can keep up.
- Source: Don't buy smart glasses, Dutch data watchdog advises — BBC
- Source: Dutch privacy watchdog AP advises consumers not to purchase smart glasses — Reuters via WHTC
AI Security: 5 Stories to Know Today
11. European Commission Confirms ARTEX AI-Assisted Cyberattack
The European Commission confirmed that hackers used ARTEX, an open-source AI penetration-testing framework, in an attack first detected on October 22. The intrusion took HR, payroll and medical reimbursement systems offline for more than 10 days. Officials found no evidence of a broad data breach and no ransom demand. It notified affected staff and is checking other systems.
CERT-EU and the European Data Protection Supervisor have been informed. ARTEX, released in July, uses commercial AI models to automate vulnerability discovery and exploitation. The FBI warned last week that some ARTEX intrusions reached privileged access in under three hours. It is the Commission's third disclosed incident this year, after Ivanti and cloud breaches. Officials have not linked them.
Why it matters: AI-powered attack tools are now breaching core government administration, where downtime directly disrupts public services.
- Source: European Commission confirms cyberattack, says hackers used AI tool — The Record
- Source: EU Commission hit by large-scale cyberattack with data leakage unknown — Euronews
12. CISA Adds Four Exploited Flaws, Including LiteLLM
CISA added four actively exploited flaws to its Known Exploited Vulnerabilities (KEV) catalog on November 18. They include Synology DSM CVE-2024-10441, LiteLLM CVE-2026-40217 and Apache Axis CVE-2026-90436. The fourth is REDCap CVE-2026-90817. Three score 9.8; LiteLLM scores 8.8. Federal agencies must patch the Synology, Axis and REDCap flaws by November 21, and LiteLLM by December 2.
The Axis flaw was exploited against Atlassian Confluence to run shell commands, Rapid7 found. LiteLLM was abused in LLMjacking campaigns that turned limited inference keys into administrative credentials. The Synology flaw was used to deploy C3RB3R ransomware, and REDCap was used to run XMRig. Google linked the LiteLLM activity to finance, retail and transport targets.
Why it matters: AI infrastructure now sits in the same exploited-vulnerability pipeline as traditional enterprise software, with the same fast patch deadlines.
- Source: CISA Adds Four Known Exploited Vulnerabilities to Catalog — CISA
- Source: CISA Warns of 4 Exploited Flaws Linked to Ransomware, Cryptomining, and LLMjacking Attacks — The Hacker News
13. PHP Patches 18 Flaws, Including Seven Critical RCE Bugs
The PHP project released updates for 18 vulnerabilities, including seven rated critical, in PHP 8.2.36, 8.3.37 and 8.4.28. PHP 8.5.13 and 8.6.0 also shipped fixes. The most serious include CVE-2026-129289, a PHP-FPM request-smuggling flaw scored 9.2, and CVE-2026-126714, a session-write race dubbed "phpAfterShell."
The critical bugs require non-default configurations or uncommon setups, PHP said. Persistent backend connections to PHP-FPM, used by some nginx, Apache and HAProxy setups, plus LSAPI, are exposed to the request-smuggling flaw. Admins can disable session upload progress or close persistent connections until they patch. Aikido published hardened container images for unsupported PHP versions back to 5.6.
Why it matters: The flaws affect the PHP runtime itself, turning routine session handling into a remotely exploitable code path.
- Source: PHP security updates fix 18 flaws, including critical RCE bugs — BleepingComputer
- Source: PHP Fixes 7 Critical Flaws, Including Bugs That Could Allow Remote Code Execution — The Hacker News
14. Apache Roller RCE Flaw Exploited Without Authentication
Apache warned that CVE-2026-129448, a critical deserialization flaw in its Roller blogging platform, is being actively exploited. The bug carries a CVSS score of 9.8, affects Roller 4.0.0 through 6.1.4, and lets an unauthenticated attacker run code through the XML-RPC endpoint before authentication. Apache did not say who is exploiting it or how many servers are affected.
The flaw is the same one tracked as CVE-2026-82384, which Apache patched in September but initially linked only to Roller 6.1.5. Users should upgrade to 6.1.6. Disabling XML-RPC in the admin interface is not enough, because older versions deserialize the request before checking that setting.
Why it matters: Niche Java applications remain useful entry points when deserialization happens before the attacker is asked to authenticate.
- Source: Critical Apache Roller RCE Flaw Under Active Exploitation, No Authentication Required — The Hacker News
- Source: Apache warns of Roller RCE vulnerability exploited in attacks — BleepingComputer
15. Nvidia Investigating Ransomware Group's Breach Claim
The Gentlemen ransomware group listed Nvidia on its data-leak site on November 18 and set a countdown to November 28. It promised "DATA INFO - SOON" but published no stolen files as proof. Nvidia said it found no indication that its own systems or networks were affected and is investigating a possible third-party partner incident.
The Gentlemen is a ransomware-as-a-service operation that offers affiliates up to 90% of ransom proceeds and has claimed more than 1,100 victims. Nvidia suffered a major breach in 2022 when Lapsus$ exposed data on 71,000 employees, and a smaller partner breach in May 2026. The group has claimed victims across manufacturing, healthcare and finance.
Why it matters: A supplier breach can expose a company's data even when its own network remains clean.
- Source: NVIDIA investigating Gentlemen ransomware breach claims — BleepingComputer
- Source: Nvidia adds possible hack to list of very bad things under investigation — The Register
More AI Stories Worth Reading
- Canva added a setting that lets users turn off its generative AI features, including image generation and Magic Expand. — The Verge
- Trump's accounts bought between $152,004 and $380,000 of Nvidia stock in September, alongside SpaceX debt and Tesla shares, as federal deals advanced. — CNBC
- OpenAI researchers would rather lose 40% of their team than go three months without coding agents, even though the tools produce "AI slop." — The Decoder
- Google's Willow processor won a quantum contextuality game 98.3% of the time, beating the 88.9% classical limit across tests up to 105 qubits. — Google Research
Related Reading on Kill The AI
- Nvidia's $115.8B quarter, Anthropic's IPO pricing, and Microsoft's agent push.
- California's OpenAI lawsuit, Huang's return offer, and the Megaspeed probe.
- Trump backing Huang, new chip export rules, and the IMF's $1 trillion warning.
- Eight weeks of AI capex, layoffs, and data-center buildout.
Methodology & Sources
Compiled November 20, 2026, using multi-source research. Sources include TechCrunch, BBC, Google, The Decoder, The Verge, the Lawyers' Committee, The Guardian, CNBC, ABC News, SiliconANGLE, Reuters via WHTC, The Record, Euronews, CISA, The Hacker News, BleepingComputer, The Register and FXStreet. All sources are free; no paywalls. Details are as reported at compilation time.
Frequently asked questions
Anthropic raised more than $81 billion and opened at $111.50, valuing it above $2 trillion. Google released Gemini 4 Neon, a faster, cheaper model with a 1 million-token context. Four Maryland students also sued over an AI alert that mistook a Doritos bag for a gun.
Anthropic priced its IPO at $83 a share, raising more than $81 billion in a record offering. The stock opened at $111.50 on Nasdaq, up 33%, valuing the company above $2 trillion. It also raised $26 billion from Google, Microsoft, Nvidia and other tech firms.
Gemini 4 Neon is Google DeepMind's fourth model release in seven weeks. It has a 1 million-token context window, costs $3 per million input tokens and $15 per million output tokens, and ranks third on the Artificial Analysis Intelligence Index at 61 points. It runs nearly twice as fast as Gemini 4 Carbon.
Four current and former Kenwood High School students sued Omnilert, Baltimore County schools and police after an AI system flagged a bag of Doritos as a gun with 90% confidence in October 2025. Armed officers handcuffed and searched them. Maryland's inspector general found more than 99% of the district's AI alerts were false positives.
CISA added four actively exploited flaws on November 18: Synology DSM CVE-2024-10441, BerriAI LiteLLM CVE-2026-40217, Apache Axis CVE-2026-90436 and REDCap CVE-2026-90817. The flaws allow remote code execution, and LiteLLM was abused to elevate inference keys. Federal agencies must patch most by November 21 and LiteLLM by December 2.
Last updated: Nov 20, 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.