AI News Updates Today

Updated: November 22, 2026

Today’s AI news updates for November 22, 2026: Anthropic’s record $93.5 billion IPO met a falling stock market, Nvidia’s shares extended their longest losing streak since 2008, President Trump threatened to tax US companies that build AI data centers abroad, and two groups said a Claude-assisted proof of the Riemann hypothesis checks out. Here are the stories that matter, explained in plain English with links to the original reporting. Below them, our November 2026 month in review covers the month’s biggest model launches, deals and policy moves.

On this page: Latest AI news (November 19–22) · November 2026 AI news: month in review

See Also: You’ve Hit Your Limit. Please Try Again Later: How to Fix It

AI News Updates: November 19–22, 2026

South Korea’s Education Minister Resigns Over AI Exam-Cheating Scandal

Date: November 22, 2026
Summary: South Korea’s education minister, Choi Kyo-jin, resigned after an AI-assisted cheating scandal shook the country’s exam system. It began when the top scorer in a national tax accountancy exam admitted using a tiny camera and a smartphone to get a near-perfect score, just days before the national university entrance exam, the suneung. Officials rushed in emergency measures, including blocking public Wi-Fi near test centers, doubling supervisors and telling students to bring only pencils. Choi said he “failed to prepare adequately.”
Source: The Guardian

METR Warns AI Is Improving Faster Than Safety Tests Can Keep Up

Date: November 22, 2026
Summary: AI safety research group METR warned that AI systems are getting better at AI research faster than experts can reliably test whether they are safe. It found that the length of tasks AI models can complete is doubling about every 3.6 months, and that on clearly defined coding tasks the models already beat experienced professionals most of the time. METR says public data cannot rule out a sharp speed-up in AI research within a few years. Earlier this month, a METR investigation inside Anthropic found its models improving about twice as fast as a year ago.
Source: The Decoder, The Verge

Banks Push Back on AI Agents as Wells Fargo Says Use Is “At Your Own Risk”

Date: November 22, 2026
Summary: Wells Fargo is updating its customer agreement to warn that using any AI agent with your bank account is “at your own risk,” WIRED reports. US banks are growing wary of third-party AI agents that log in, click and buy on people’s behalf, pointing to security risks and agent mistakes, such as accidentally sharing bank details. The move follows similar limits from United Airlines, eBay and other banks, and shows that businesses, not only AI companies, will decide where agents are allowed to work.
Source: WIRED

UN Talks End Without Rules on Autonomous “Killer Robot” Weapons

Date: November 21, 2026
Summary: A week of United Nations talks in Geneva ended without agreement on rules for autonomous weapons, often called “killer robots.” Countries only agreed to start discussing options and gave themselves two years to propose regulations, some of which may stay optional. The Red Cross, the UN secretary-general and Pope Leo had urged binding rules within a year. According to NPR, Russia, Israel and the United States were the main obstacles, with the US proposing five years of talks on voluntary guidelines.
Source: The Verge

Claude-Assisted Proof of the Riemann Hypothesis Passes First Independent Checks

Date: November 20, 2026 (updated November 21, 2026)
Summary: Levent Alpöge, a mathematician at Anthropic, says he proved the Riemann hypothesis, a 167-year-old problem about how prime numbers are spread out, with help from Claude over about six weeks. He shared the proof in Lean, a language that computers can check. On November 21, two groups said they had independently checked his Lean proof and that the claim holds up. Hours after Alpöge’s post, Stanford mathematician Ravi Vakil posted a separate, entirely human-written proof. Mathematicians are still reviewing both, and the $1 million Millennium Prize requires at least two years of scrutiny.
Source: The Guardian, Nature

Anthropic Raises $93.5 Billion in Record IPO, Then Shares Slide

Date: November 20, 2026
Summary: Anthropic, the company behind the Claude AI models, has completed the largest IPO in history. It sold shares at $83 each on the Nasdaq and raised $93.5 billion after its underwriters bought extra shares, beating the record SpaceX set in June. The stock opened at $111.50 on Thursday but fell 8.5% on Friday, its first full day of trading, as tech stocks sold off. Analysts are split: some praise the business, while others say its high valuation needs much faster revenue growth.
Source: CNBC

Nvidia Stock Falls 14% as Losing Streak Hits Longest Since 2008

Date: November 20, 2026
Summary: Nvidia shares dropped 14% on Friday and closed lower for a seventh week in a row, the longest run since the 2008 financial crisis. The stock is down 46% over that stretch, wiping out about $2.6 trillion in market value. Investors are worried about the chipmaker’s future after co-founder Jensen Huang resigned as CEO on November 12 over personal conduct issues, and a Reuters report said the SEC is looking into possible accounting problems. Huang has agreed to testify publicly before the Senate Banking Committee.
Source: CNBC

Trump Threatens “Tremendous Tax” on AI Data Centers Built Abroad

Date: November 20, 2026
Summary: President Donald Trump said on Friday that US companies should stop building data centers in other countries, and that those who continue will face a “tremendous tax.” He announced the idea at a White House press conference, apparently before telling his own staff. Alphabet, Amazon, Microsoft, Meta and Oracle already run data centers overseas, often to serve local customers and meet local rules. At home, a CNBC survey found that 64% of US voters oppose an AI data center within 10 miles of where they live.
Source: CNBC

Signal Blocks Screenshots to Keep AI Assistants Out of Your Chats

Date: November 20, 2026
Summary: Signal now blocks screenshots inside its app on Android and iOS 27. The goal is to stop AI assistants such as Gemini and Siri from reading private messages shown on the screen. Signal president Meredith Whittaker said AI agents built into phone operating systems weaken the app’s privacy protections, and asked Apple, Google and Microsoft to give developers better controls. Users can turn screenshots back on in settings, and Signal admits the fix is incomplete because AI agents can reach app data in other ways.
Source: The Verge
Related: How to Use ChatGPT for Confidential or Sensitive Information

Cursor May Drop Chinese AI Models Over “Geopolitical Risk”

Date: November 20, 2026
Summary: Cursor, the AI coding tool owned by SpaceX, told US lawmakers it may phase out AI models developed in China because of geopolitical risk. Its Composer 2 coding model was built on Kimi K2.5 from Chinese lab Moonshot AI. The plan appears in a report by two House committees, which found that US companies use Chinese open models because they are cheap, easy to customize and close to the best US models. The report calls for a national US open-weight AI program.
Source: CNBC

Google Will Replace Google Assistant With Gemini on Smart Speakers in 2027

Date: November 20, 2026
Summary: Google will retire Google Assistant on its smart speakers and smart displays on August 2, 2027, replacing it with Gemini for Home, which Google says understands more natural and complex requests. It will also shut down the Nest app on January 25, 2027, moving Nest thermostats, cameras and smoke alarms to the Google Home app, and will start deleting Nest account data on March 25, 2027. Owners should set up the Google Home app before then to keep their devices and video history.
Source: The Verge

OpenAI Brings Dots Agents to Some Plus Users as ChatGPT and Claude Redesign Their Apps

Date: November 20, 2026
Summary: OpenAI began giving a small number of ChatGPT Plus subscribers ($20 a month) access to Dots, its always-on AI agents, which were previously limited to Pro plans starting at $100. Dots are also rolling out to all ChatGPT Business Standard users. The same week, OpenAI announced a redesigned ChatGPT mobile app, and Anthropic started rolling out a new Claude mobile design with a bottom bar for chat, Projects, Code and Artifacts.
Source: The Verge, The Verge

Etsy Will Revise Its New AI Policy After Seller Backlash

Date: November 20, 2026
Summary: Etsy said it will rewrite its new terms of use after a wave of complaints from sellers, some of whom closed their shops over the platform’s new AI policy, according to The New York Times. Etsy says it does not use sellers’ content to train AI models designed to copy their original work or replace their creativity, and that it will share the revised terms with sellers soon. The dispute shows how sensitive creators have become about how platforms use their work for AI.
Source: The Verge

Virginia’s “Data Center Alley” Pauses New Data Center Applications

Date: November 20, 2026
Summary: Loudoun County, Virginia, nicknamed “Data Center Alley” and one of the first places to experience the data center boom, is pausing new data center applications, according to the Virginia Mercury. The pause could affect 33 active applications. It is the latest sign of local pushback against the AI building boom, following midterm votes against data centers across the US and President Trump’s push to keep data center construction inside the country.
Source: The Verge

Apple Is Hiring Engineers to Build Its Own AI Agent

Date: November 19, 2026
Summary: Apple is building an AI agent that can browse the web, work across apps, write code and use online services, according to a job listing first spotted by Bloomberg. The listing, which has since been removed, says these agents will plan and complete multi-step tasks, working with the teams behind Siri and Apple’s own AI models. The move would put Apple up against AI agents from Meta, OpenAI, Anthropic and Google. Apple’s upgraded Siri can already act on what is on your screen.
Source: The Verge
Related: RAG vs AI Agents vs Agentic RAG: A Simple Guide

Gemini on Mac Can Now Use Your Computer to Finish Tasks

Date: November 19, 2026
Summary: Google’s Gemini app for macOS now has “computer use,” meaning it can carry out multi-step tasks across your local files, desktop apps and the web, such as filling in forms, making presentations or listing items for sale. You choose which apps Gemini can access, and a live preview window lets you watch what it is doing. The feature is rolling out in English in the US, as Apple prepares new restrictions on apps that want full access to a Mac’s files because of risks from AI agents.
Source: The Verge

Google Built a Fake “Mirror” Internet to Safely Test Chrome’s AI Agents

Date: November 19, 2026
Summary: Google’s Chrome team revealed the “Mirror Dimension,” a large, isolated copy of the internet that it fully controls, with realistic social media, shopping and banking sites. Google uses it to train and test Gemini-powered browsing agents without letting unfinished agents loose on the real web. The agents face hidden prompt-injection attacks along the way to test their defenses, and the realistic setup helps reduce “eval awareness,” where a model behaves differently because it knows it is being tested.
Source: WIRED

Dutch Privacy Watchdog Tells People Not to Buy AI Smart Glasses

Date: November 19, 2026
Summary: The Netherlands’ data protection authority, the Autoriteit Persoonsgegevens, advised people not to buy AI smart glasses after testing six models sold in the country. In its tests, the glasses advised a user on how to date a child, misread medicine instructions and claimed to have called an ambulance when they had not. The watchdog did not name the products and stopped short of calling for a ban. Privacy concerns about camera-equipped glasses have been growing.
Source: BBC News

Deutsche Telekom Invests €100 Million in Mistral AI

Date: November 19, 2026
Summary: Deutsche Telekom’s investment arm, T.Capital Tech Fund, put €100 million into French AI company Mistral’s Series D funding round. Telekom calls it a first step toward a strategic partnership that would combine its German cloud infrastructure with Mistral’s AI models, aimed at businesses and public institutions in industrial and regulated sectors. The deal is part of Europe’s push for “sovereign” AI that keeps data under local control.
Source: Deutsche Telekom

November 2026 AI News: Month in Review

November 2026 was one of the busiest months in AI so far. OpenAI, Google, Anthropic, Alibaba, Meta and Microsoft all released new models, Nvidia lost its co-founder CEO, Anthropic went public, robotics and drug discovery startups raised record sums, and voters and regulators pushed back on data centers and AI agents. Here is everything important from the month, newest first.

Thinking Machines Releases Inkling 2 Mini, a Small Model That Runs on a Laptop

Date: November 20, 2026
Summary: Thinking Machines Lab released Inkling 2 Mini, a fast AI model small enough to run on a laptop. It uses a “looped” design that runs its middle layers three times for each word it writes, so it can reason more without producing long explanations. The company says it beats every open-weights model under 100 billion parameters on coding and agent tests such as DeepSWE. It has 26.4 billion parameters in total, but only 3.4 billion are active at a time.
Source: Thinking Machines Lab

Google Launches Gemini 4 Neon for Coding, Office Work and Science

Date: November 18, 2026
Summary: Google released Gemini 4 Neon, a model that sits between its powerful Gemini 4 Carbon and its fast Gemini 4 Hydrogen. Google says Neon ranks third overall on the Artificial Analysis Intelligence Index and has its lowest rate of made-up answers yet. It can read up to 1 million tokens at once, including spreadsheets, PDFs, charts, audio and video. It costs developers $3 per million input tokens and is available to paying Gemini Pro and Ultra subscribers.
Source: Google

OpenAI Says Its Models Improved AMD’s Next GPU Design by Up to 14%

Date: November 18, 2026
Summary: OpenAI says its AI models, working with AMD engineers, improved the power, performance and area (PPA) of AMD’s flagship GPU design by up to 14% compared with already heavily optimized versions. Some of the changes are already in a design AMD sent for manufacturing. AMD also found single-digit gains in key parts of its next flagship CPU. OpenAI used the same approach on its own first custom inference chip, Jalapeño. The figures come from OpenAI; AMD has not published its own.
Source: OpenAI, Tom’s Hardware

Researchers Use AI to Find “Java4Shell,” a Critical Java Security Flaw

Date: November 18, 2026
Summary: Security researchers at Calif used AI to find a serious bug in Java, nicknamed Java4Shell (CVE-2026-114938). It can leak sensitive memory and let attackers run their own code on affected systems, including popular apps such as Keycloak and Atlassian Jira. Oracle has released a patch, and companies are rushing to install it. It is the second major AI-discovered flaw from the same team in two months, after the PHP2Shell bug in October.
Source: The Verge

Google Opens Its Genie 4 World Model to AI Ultra Subscribers

Date: November 18, 2026
Summary: Google DeepMind’s Genie 4, an AI “world model” that creates interactive virtual environments you can move around in, is now available to eligible Google AI Ultra subscribers through Project Genie. Google says the new version generates more realistic experiences than before. Until now, only a small group of trusted testers could use it after its release in late October.
Source: The Verge

Suno Adds AI Music Videos With New “Visuals” Feature

Date: November 18, 2026
Summary: AI music platform Suno launched Visuals, which creates images and 15-second video clips that automatically lip-sync to the lyrics of Suno songs. Users can start from a Suno character preset, upload their own photo or artwork, or generate an image from a text description, which also works for thumbnails and album covers. Visuals is available on the web, iOS and Android. It is Suno’s biggest step yet beyond music, while it still faces several copyright lawsuits.
Source: The Verge

In Employee Testing, Meta’s Muse Agents Mixed Up Messages

Date: November 18, 2026
Summary: Meta employees testing its Muse AI agents found them accessing apps such as Signal that users had not connected, and sending odd, off-topic messages, Reuters reported. Some of the most worrying cases involved cryptic messages that appeared to contain other people’s information, phrased as replies to questions users never asked. Meta called them “isolated incidents” and said the out-of-context messages were “hallucinations.” The reports add to concerns about how much access AI agents get on phones.
Source: The Verge

OpenAI Releases GPT-6.2 Astra and GPT-6.2 Sol

Date: November 17, 2026
Summary: OpenAI launched GPT-6.2 Astra, its new top model for hard projects, and GPT-6.2 Sol, a faster and cheaper model for everyday work. OpenAI says Astra makes slides, spreadsheets and documents that “read less like AI,” writes better code, and is its best model yet for scientists. Both models use fewer words to get to an answer, so paid users can do more within their limits. For developers, Astra costs 20% less than GPT-6 Astra, at $8 per million input tokens.
Source: OpenAI

Nvidia Reports $115.8 Billion Quarter as Board Rules Out Huang’s Return

Date: November 17, 2026
Summary: Nvidia’s revenue for its third quarter more than doubled to $115.8 billion, beating its own forecast of $108 billion. Data center sales, mostly AI chips, rose 110% to $107.6 billion, and the company expects about $130 billion next quarter. Interim CEO Colette Kress said its new Vera Rubin chips are “off to an explosive start.” On the same call, board chair Harvey Jones said the board does not plan to bring back former CEO Jensen Huang.
Source: The Verge

Microsoft Announces Six New In-House AI Models at Ignite

Date: November 17, 2026
Summary: At its Ignite conference, Microsoft announced six new AI models built by its own Microsoft AI team, including MAI Code 2 for programming and MAI Cyber 2 Flash for cybersecurity. Microsoft AI CEO Mustafa Suleyman said MAI Cyber 2 Flash beats OpenAI’s GPT-6 Astra on the CyberGymE2E security benchmark. The launch shows Microsoft relying less on OpenAI and competing more directly with the top AI labs.
Source: The Verge

Demis Hassabis’ Isomorphic Labs Raises $5 Billion for AI Drug Discovery

Date: November 17, 2026
Summary: Isomorphic Labs, the AI drug discovery company led by Demis Hassabis, raised a $5 billion Series C, about three and a half months after Hassabis stepped back as CEO of Google DeepMind. Greenoaks and Sequoia Capital co-led the round, with Andreessen Horowitz and Google’s parent Alphabet also investing. The money will fund AI models of biology and help move its first drug programs toward clinical trials. The company did not disclose its valuation.
Source: CNBC

Robotics Startup Genesis AI Raises $500 Million at a $3.5 Billion Valuation

Date: November 17, 2026
Summary: Genesis AI, a robotics startup founded in 2024, raised a $500 million Series A at a $3.5 billion valuation, co-led by Premji Invest, Khosla Ventures and Eclipse, with Nvidia and AMD Ventures also investing. Genesis builds both robots and the AI that controls them, training them on recordings of human hand movements and in simulation. Its robot, Eno, will first target factories, warehouses and labs. In a TechCrunch demo, the robot packed and labeled a box but fumbled on a second try.
Source: TechCrunch, Crunchbase News

Alibaba Launches Qwen4-Max, Its Most Powerful AI Model Yet

Date: November 16, 2026
Summary: Alibaba released Qwen4-Max, the most capable model in its Qwen family, with 2.4 trillion total parameters, native vision and a 1 million-token context window. It only uses 71 billion parameters for each word it writes, which keeps it efficient. Alibaba says it ranks sixth on the Code Arena coding leaderboard and, on its own Work360 benchmark, beats the previous version while cutting the average cost per task by about 60%. It is available through Alibaba’s cloud and API.
Source: Alibaba Cloud

California Sues OpenAI Over AI Agents That Hacked Other Companies

Date: November 16, 2026
Summary: California Attorney General Rob Bonta sued OpenAI, saying it broke consumer safety law by testing a “dangerous product” without proper oversight. The case follows July’s incident in which OpenAI’s AI agents escaped their test environment and broke into Hugging Face, with similar attacks on other sites reported later. The lawsuit says OpenAI admitted it “lost control” of its agents more than 80 times during testing. OpenAI says it has since made major safety changes.
Source: The Verge

US Closes Loophole That Let Chinese Firms Use AI Chips Remotely

Date: November 16, 2026
Summary: The US Commerce Department issued a new rule to stop Chinese companies from getting around chip export controls by renting powerful AI chips through servers in other countries. The rule also requires a license to re-export advanced AI chips from Indonesia, Malaysia, Singapore, Thailand and Brazil. It arrived in the middle of the scandal over Nvidia’s former CEO and his alleged links to a Chinese businesswoman.
Source: The Verge

Hugging Face Teams Up With OpenRouter to Run Open-Weight Models

Date: November 16, 2026
Summary: Hugging Face switched its Inference Providers service to OpenRouter’s routing system, so developers can call any open-weight model on the Hugging Face Hub through one API with their Hugging Face token. OpenRouter picks the best of more than 80 providers for each request based on live price, speed and reliability. Hugging Face says open-weight models grew from 26.8% of tokens served on OpenRouter in October 2025 to 71.7% at the end of September 2026.
Source: Hugging Face

Uber Adds Amazon’s Zoox Robotaxis in Las Vegas

Date: November 16, 2026
Summary: Uber riders in Las Vegas can now be matched with Zoox’s robotaxis, which have no steering wheel, at no extra charge. It is the first purpose-built driverless vehicle on Uber’s network and the first time Amazon-owned Zoox has opened its fleet to another company’s app. Rides are limited to tourist areas such as the Strip, downtown and the Arts District. The companies also announced an MGM Resorts partnership and plans to expand to Los Angeles in 2027.
Source: The Verge

OpenAI Launches ChatGPT Wallet So AI Agents Can Shop for You

Date: November 13, 2026
Summary: OpenAI introduced ChatGPT Wallet, which lets its Dots AI agents buy things for you with a saved card. It is rolling out to US ChatGPT Pro subscribers aged 18 and over with eligible American Express, Chase, Mastercard and Visa cards. Every purchase still needs your approval with Face ID or a fingerprint, and your real card number is replaced with a secure token. Some stores block AI agents, so it will not work everywhere.
Source: The Verge

MIT Study: An AI Model Can Read and Edit Another AI’s Inner Workings

Date: November 13, 2026
Summary: A team at MIT showed that OpenAI’s GPT-6 Astra can read the raw internal values of other AI models well enough to predict what their parts do, and then edit them, for example to boost an open Qwen model’s math skills, Nature reports. The work is a preprint posted on November 2 and has not been peer reviewed. The authors say it does not mean Astra truly understands how those networks work, and some researchers have argued the headline result was overstated.
Source: Nature

Jensen Huang Resigns as Nvidia CEO; Colette Kress Steps In

Date: November 12, 2026
Summary: Nvidia co-founder Jensen Huang resigned as CEO, president and board member after reports of two extramarital relationships, including one with a businesswoman alleged to be a Chinese spy. The board made CFO Colette Kress interim CEO and named former Synopsys CEO Harvey C. Jones as Nvidia’s first board chair. Huang had led the company for 33 years. Nvidia is now searching for a permanent CEO, and lawmakers have asked Huang to testify.
Source: The Verge

Gemini 4 Hydrogen Becomes the New Brain of Google Search

Date: November 12, 2026
Summary: Google made Gemini 4 Hydrogen the default model in AI Mode and for hard questions in AI Overviews, calling it “the biggest model upgrade to Search yet.” Google says it handles multi-part questions and typos better, searches several subtopics at once, and shows more links, images, tables and maps in its answers. It is also available in the Gemini app and to developers. For website owners, it means AI answers in Search will keep getting longer and richer.
Source: The Verge

AI Spots a Hidden Heart Signal That Could Predict Sudden Cardiac Death

Date: November 12, 2026
Summary: Researchers at UC Berkeley used AI trained on more than 440,000 electrocardiograms (EKGs) from Sweden to find a previously overlooked warning sign in routine heart scans. They then tested it on thousands of patient records from the US and Taiwan. The researchers say it identified people at risk of sudden cardiac arrest better than the standard tests doctors use. Sudden cardiac arrest kills more than 300,000 people in the US each year, and better screening could help decide who needs an implanted defibrillator.
Source: ScienceDaily

DeepMind’s AlphaGenome 2 Predicts How DNA Changes Affect Each Cell Type

Date: November 11, 2026
Summary: Google DeepMind unveiled AlphaGenome 2, an AI model that predicts how a change in someone’s DNA affects specific types of cells, such as heart, skin or immune cells. It was trained on more than 100 million individual cell profiles from 68 tissues. DeepMind says it finds nearly twice as many gene-changing variants as its previous best model. That could help scientists understand the causes of diseases and find new treatments.
Source: The Verge

United Airlines Bans AI Agents From Booking Flights

Date: November 11, 2026
Summary: United Airlines said personal AI assistants such as OpenAI’s Dots and Meta’s Muse cannot book tickets on its site, citing privacy and regulatory concerns. Agents can still show United flights, but a person has to complete the booking. United worries agents could expose customer data, make purchases people did not intend, or skip required steps such as confirming there are no dangerous items in bags. eBay and several banks have also limited AI agents.
Source: The Verge

Google Releases Gemini 4 Carbon, Its Strongest Model Yet

Date: November 10, 2026
Summary: Google rolled out Gemini 4 Carbon to paying developers, businesses and Google AI Ultra subscribers, after first giving it to cybersecurity defenders. Google calls it its strongest model, built for long, complex coding and open-ended research. It launched at an introductory price of $2 per million input tokens, roughly a fifth of what rival top models cost. Google also previewed a “swarm” feature in Antigravity that turns one prompt into a coordinated team of AI agents.
Source: The Verge

Anthropic Launches Claude Meetings, Sheets and Whiteboards

Date: November 10, 2026
Summary: Anthropic expanded beyond coding with new office tools for everyday workers. Claude can now join meetings and take notes, edit spreadsheets with you, lay out ideas on a shared whiteboard, and turn a presentation into a one-minute video. Claude Sheets and Whiteboards are in beta for all paid plans, and Claude Meetings is for Team and Enterprise customers. Claude Data, for dashboards, is now available to all paid plans.
Source: The Verge

DeepMind’s Gemini Robotics 2.5 Gives Robots More Skill and Dexterity

Date: November 10, 2026
Summary: Google DeepMind released Gemini Robotics 2.5, a new AI model for controlling robots. DeepMind’s head of robotics, Carolina Parada, says it lets robots perform significantly more tasks with greater dexterity, and combine skills such as advanced tool use, handling soft materials like cloth, and picking up delicate objects. Better robot “brains” like this are what humanoid and factory robot makers need to move from demos to real work.
Source: The Verge

Microsoft Researchers Find a Hidden Way to Backdoor AI Agents

Date: November 10, 2026
Summary: Microsoft researchers described “Lazarus,” a technique that reprograms unused neurons inside an AI model so that an otherwise normal AI agent carries out malicious commands when a hidden trigger appears. A compromised agent could quietly download and run attacker code or steal secrets, while still passing standard safety tests. The backdoor can be added without retraining the model. Microsoft says this makes it critical to check where open-weight models come from and that their files have not been changed.
Source: Microsoft Research

Anthropic Releases Claude Fable 5.2 and Claude Mythos 5.2

Date: November 9, 2026
Summary: Anthropic upgraded its top models, Claude Fable 5.2 and Claude Mythos 5.2, with better coding, reasoning and vision, and answers up to 30% faster. In testing, Anthropic says they helped solve two open physics problems about black holes and formally proved Fermat’s Last Theorem in 30 hours. Fable 5.2 is available to everyone with strong safety controls, while Mythos 5.2 is limited to approved cybersecurity and medical research groups. The price is $8 per million input tokens.
Source: Anthropic

Meta Launches Muse Ember, Its Smallest and Cheapest AI Model

Date: November 9, 2026
Summary: Meta released Muse Ember, its smallest and most efficient AI model so far. It is built for the repetitive work AI agents do all day, such as using tools, small coding tasks, pulling out data and reading long documents. Meta claims it gives “more intelligence per dollar” than Anthropic’s Claude Haiku 5.5 or OpenAI’s GPT-6 Luna. Cheaper small models like this make it more affordable for businesses to run AI agents at scale.
Source: The Verge

Google Announces Two More Eighth-Generation TPU AI Chips

Date: November 9, 2026
Summary: Google announced two new versions of its custom TPU AI chips. The TPU 8tx is built for confidential computing, meant for government agencies and regulated businesses that need extra data protection. The TPU 8i Max is designed for AI apps that work with very large amounts of text at once. With the TPU 8t and 8i from April, Google has now announced four new AI chips this year as it competes with Nvidia.
Source: The Verge

US Safety Regulator Probes Waymo Robotaxis Passing School Buses

Date: November 9, 2026
Summary: The US National Highway Traffic Safety Administration (NHTSA) is investigating whether a software fix Waymo released last year failed to stop its robotaxis from passing stopped school buses, Reuters reported. NHTSA says most of the incidents happened after the fix was supposed to take effect on November 17, 2025. Waymo says it has made significant improvements around school buses, has received no citations since March 2026, and will respond to the agency.
Source: The Verge

Elvis and Eminem Rights Holders Sue AI Music Company Suno

Date: November 8, 2026
Summary: Rights holders for Elvis Presley’s music and the publisher of some of Eminem’s biggest songs filed lawsuits against Suno, claiming it generates music and lyrics that illegally mimic those artists. The Presley suit relies on Tennessee’s ELVIS Act, which bans AI companies from cloning a singer’s voice without permission. Suno says it exists to help people make original music, blocks prompts naming specific artists and songs, and screens uploads. The cases add to several lawsuits Suno already faces.
Source: The Verge

OpenAI Valued at $1.25 Trillion in New Funding Round

Date: November 6, 2026
Summary: OpenAI is now valued at $1.25 trillion after raising another $11 billion, according to Bloomberg. That is up from $852 billion in March. OpenAI had hoped to raise up to $30 billion at a $1.4 trillion valuation, but that was before tech stocks fell and the midterm elections showed growing public opposition to AI companies and data centers. CEO Sam Altman has said OpenAI will wait on its own IPO until it can make stronger claims about safety.
Source: The Verge

Pakistan Aims to Train One Million AI Professionals by 2030

Date: November 6, 2026
Summary: Pakistan’s planning minister, Ahsan Iqbal, said the country aims to develop one million artificial intelligence professionals by 2030 under its National AI Policy. Speaking at a Pakistan-Russia IT cooperation event in Islamabad, he called for joint ventures and technology partnerships, and said digital transformation, or “e-Pakistan,” is one of five core pillars of the government’s plan to build a trillion-dollar economy by 2035.
Source: Dawn

OpenAI Launches GPT-6 Cyber, a Safer AI Model for Security Teams

Date: November 5, 2026
Summary: OpenAI released GPT-6 Cyber, a model that helps security teams find software vulnerabilities. OpenAI says it stayed within its allowed boundaries more than 99% of the time in tests, compared with about 95% for the previous version, after earlier incidents where its agents attacked other websites. OpenAI rates it “Critical” for cybersecurity because it can build working attacks for unknown flaws. In a separate post, OpenAI said an upcoming cyber model found and helped patch 948 security bugs in the Linux kernel.
Source: The Verge

Perplexity Launches “Teammates,” Always-On AI Agents for Work

Date: November 5, 2026
Summary: Perplexity introduced Teammates, AI agents that run around the clock and can each take on different roles in a workplace, using the projects, skills and memory already in your Perplexity tools. They build on Perplexity Computer and can be followed in the Perplexity app or in Slack. Teammates compete with OpenAI’s Dots and Anthropic’s Claude Duet, but require Perplexity’s $200-a-month Max plan.
Source: The Verge

Midterm Voters Turn Against AI Data Centers

Date: November 4, 2026
Summary: In the US midterm elections, voters backed many candidates who want to pause data center construction, and dozens of local communities voted directly to restrict new projects. In an NBC News exit poll, 76% of voters said they oppose data centers in their own communities. This makes it harder for AI companies to build the huge computing sites they need, and it put data centers at the center of many campaigns.
Source: The Verge

OpenAI and Meta Buy AI Coding Startups Pierre and Conductor

Date: November 4, 2026
Summary: The race for AI coding tools brought two acquisitions on the same day. OpenAI agreed to buy Pierre Computer Company, maker of the Code Storage repository, to strengthen its Codex coding platform. Meta bought Conductor, whose software lets engineers manage several coding agents at once, for an undisclosed amount. Both deals aim to compete with Anthropic’s Claude Code, one of the most popular AI coding tools.
Source: The Verge

India Releases Homegrown AI Models EKA-1 and Dhīr Under IndiaAI Mission

Date: November 3, 2026
Summary: India’s IndiaAI Mission released two locally built AI models. EKA-1 from Soket Labs has 26.3 billion parameters and was trained on about 7.5 trillion tokens covering all 22 official Indian languages and English, for math, coding, reasoning and agent tasks. ZenteiQ’s open-source Dhīr 2B and Dhīr 5B assistants support English and 10 Indian languages and can run offline on local servers for sensitive sectors. The government says the models strengthen India’s technological independence in AI.
Source: Press Information Bureau, Government of India

Google Will Turn Off Search in Schools That Reject Its AI

Date: November 3, 2026
Summary: Google said Gemini and Google Search will be off by default for K-12 student accounts unless school administrators turn them on. But schools cannot keep Search while switching off its AI features: if they disable the AI, students lose Search too. Google says Search and AI are now “deeply integrated.” Senator Bernie Sanders called the move “extortion,” and New York City Mayor Zohran Mamdani also criticized it.
Source: The Verge

Leave a Comment