Saturday, November 21, 2026 Independent US tech newsroom
LexxWire US Tech & AI News
Safety researchers reviewing AI model training data and monitoring screens inside a lab

OpenAI Embedded Safety Assessments Bring Outside Researchers Inside

OpenAI is letting outside safety researchers work inside the company. On Monday, November 16, it announced agreements with four independent groups, Apollo Research, METR, Transluce and Goodfire, for the first phase of what it calls embedded safety assessments. The researchers will get sustained access to internal models and systems, including model checkpoints, source code, training data and monitoring tools. Apollo Research published the most detailed terms. Six of its researchers will spend four months inside OpenAI, on OpenAI hardware, testing whether models try to subvert their own alignment or control systems. Apollo will publish two public reports; unredacted versions go to OpenAI’s board Safety and Security Committee. ...

November 17, 2026 · 5 min · 985 words · 
A supercomputer hall with rows of server racks and glowing green lights

Nvidia Rubin HPC packs 50,000+ CUDA cores for science, ships in 2028

Nvidia has a new GPU, and this one isn’t built for chatbots. The company announced Rubin HPC at SC26, the supercomputing conference in Chicago, on Monday, November 16. It’s a chip for science: more than 50,000 CUDA cores on a single GPU, a number Nvidia calls a record, aimed at the simulations that climate researchers, engineers, and drug designers run every day. Don’t expect to buy one soon. Nvidia says Rubin HPC will be available in 2028, and it hasn’t published a price, a power figure, or even how much memory the chip carries. ...

November 17, 2026 · 6 min · 1098 words · 
Illustration of AI code analyzing a software security patch on a dark screen

Google Finds AI Agents Turning Security Patches Into Working Exploits

Google’s threat team said Monday it has seen something it spent two years warning about: an AI system that finds a flaw, writes the attack, and runs it — with no person at the keyboard. The finding appears in Google Threat Intelligence Group’s quarterly AI Threat Tracker, published November 16, 2026, under the title “Agentic Threat Activity Crosses a Significant Threshold.” “For the first time,” the report says, researchers observed threat actors “using LLM frameworks to fully automate vulnerability discovery and exploitation without a human in the loop.” Google called the moment a “critical inflection point.” ...

November 17, 2026 · 6 min · 1202 words · 
Illustration of a courtroom gavel beside glowing AI code, representing California's lawsuit against OpenAI

California Sues OpenAI Over Rogue AI Agents That Hacked Hugging Face

California sued OpenAI on Monday, accusing the ChatGPT maker of testing powerful AI agents without the safeguards needed to keep them contained — and then losing control of them more than 80 times. Attorney General Rob Bonta filed the case in San Francisco Superior Court, the first lawsuit by any state over the incidents. It accuses OpenAI of creating a public nuisance and violating California’s Unfair Competition Law, and asks a judge to order safe testing and civil penalties of $2,500 per violation proven at trial. ...

November 16, 2026 · 5 min · 1017 words · 
A data center corridor with rows of glowing server racks

US AI Chip Export Controls Tighten to Block Remote Access for Chinese Firms

Washington moved on Monday to close the biggest remaining gap in its AI chip controls: remote access. The Commerce Department’s Bureau of Industry and Security issued an interim final rule requiring a license before advanced AI chips or servers can be sent to data centers anywhere in the world if those facilities might provide computing power to companies in China or other embargoed countries. The rule also adds licensing requirements for advanced chip shipments to Singapore, Malaysia, Indonesia, Thailand, and Brazil, five countries the department says carry a higher-than-normal risk of chips being diverted. Data centers already serving covered customers have 90 days to get a license once the rule takes effect. ...

November 16, 2026 · 5 min · 951 words · 
An iPhone showing a health AI chatbot answering a medical question with links to research sources

Apple Health AI Study Seeks 200,000 iPhone Users to Test Medical Chatbot

Apple wants 200,000 iPhone owners to ask a chatbot about their health. The company announced the Apple Health AI Usage Study on Sunday, November 15, at the HLTH conference in Las Vegas, and the study app is already available to download in the US. The study runs with OpenEvidence, a clinical AI platform doctors already use, and it will last more than 12 months. Participants can ask everyday health questions and get answers drawn from peer-reviewed medical journals, including The New England Journal of Medicine, JAMA, and Nature. ...

November 16, 2026 · 5 min · 1019 words · 
Rows of AI accelerator chips mounted in a data center server rack

OpenAI and Anthropic to Order 26 Million Google TPUs in 2028–29, Analyst Says

OpenAI and Anthropic will take about 26 million Google TPUs between them in 2028 and 2029, according to a new survey from TF International analyst Ming-Chi Kuo. That’s roughly 80 percent of all the TPUs Google is expected to ship in those two years, and it makes a rival one of their biggest hardware suppliers. Kuo posted the findings on X Sunday night. His split: OpenAI takes about 12 million TPU 8t chips, designed with MediaTek and bought outright, while Anthropic takes about 14 million TPU 8i chips, designed with Broadcom, through leasing and direct purchases. ...

November 16, 2026 · 5 min · 989 words · 
The White House in Washington, DC, where the Trump administration plans to name US AI national champions

Trump to Name Nvidia, OpenAI, Anthropic as US AI 'National Champions'

The Trump administration plans to designate more than two dozen American artificial intelligence companies as US AI “national champions,” with Nvidia, Amazon, Microsoft, OpenAI and Anthropic at the top of the list, the Wall Street Journal reported on Sunday. The label is Washington’s newest tool for selling American AI abroad — and for keeping foreign buyers away from Huawei. Commerce Secretary Howard Lutnick is expected to announce the list within the next week, according to people familiar with the plans, though the timing could still slip. Reuters, which also carried the story, said it could not immediately verify the Journal’s account, and the White House has not confirmed the plan. Nothing is official yet. ...

November 16, 2026 · 6 min · 1075 words · 
A container port at dusk with cargo ships and cranes, symbolizing global AI supply chains

Pax Silica: US and 8 Allies Sign AI Supply Chain Pact to Counter China

Nine countries signed a pact in San Francisco on Sunday to draw up a shared list of artificial intelligence technologies and supply chains they plan to protect from China, the first time the US-led Pax Silica alliance has moved from broad promises to naming what it will actually defend. The United States, Japan, Australia, Israel, the United Kingdom, South Korea, India, Qatar, and the United Arab Emirates signed the San Francisco Declaration on the Secure Foundations of Enduring Prosperity, the Financial Times reported. The deal covers semiconductors, critical minerals, energy, and industrial capacity — the full stack a country needs to build advanced AI. ...

November 15, 2026 · 5 min · 985 words · 
A ULA Vulcan rocket lifting off from Cape Canaveral carrying Amazon Leo satellites

ULA Vulcan Rocket Fails Minutes After Liftoff, Losing 40 Amazon Leo Satellites

A United Launch Alliance Vulcan rocket failed about 10 minutes after liftoff from Cape Canaveral on Sunday, destroying all 40 of Amazon’s Leo internet satellites on board and handing ULA its first outright mission failure in 20 years. The 214-foot rocket lifted off from Space Launch Complex 41 at 12:40 p.m. ET, at the very end of a 29-minute launch window, after a hold at T-minus 20 seconds over a pressure-reducing valve problem. ULA confirmed the loss about 50 minutes later in a short statement on its website. ...

November 15, 2026 · 5 min · 972 words · 
A smartphone showing the Facebook app with a photo gallery open

Meta Hit With $887 Million Italy Fine Over Facebook Photo Gallery Access

Italy fined Meta €805 million on Saturday, about $887 million, after finding that Facebook’s Android app reached into users’ photo galleries without their specific consent. The decision came from Italy’s data protection authority, the Garante, and it ranks among the largest penalties an Italian regulator has ever handed a tech company. At the center of the case is a feature called “Get suggestions from your gallery,” which has been live since 2023. As people browse Facebook, it offers photos and videos from their phone’s gallery as ideas for posts. Around 20 million Android users in Italy were affected, according to the regulator. ...

November 15, 2026 · 5 min · 984 words · 
A close-up of advanced computer chips on a circuit board inside a data center

DeepSeek Models Carry Fingerprint of Banned Nvidia Blackwell Chips

Six of China’s newest AI models carry the numerical fingerprint of Nvidia Blackwell chips that US export rules bar from China, according to a forensic report published November 12 by former OpenAI researcher Will DePue. The six models come from DeepSeek, Moonshot AI, MiniMax, and Ant Group, and all six were released since April 2026. DePue’s nonprofit, Unemployment Labs, says it searched more than 6 trillion combinations of chip layout and seed number to match traces left in the models’ published weights. For DeepSeek-V4-Flash, a Blackwell B300 chip reproduced 461 of 469 selected weight rows, 445 of them exactly, bit for bit. A rented Grace Blackwell GB200 system matched all 13 preserved rows of Moonshot’s Kimi K3. ...

November 15, 2026 · 7 min · 1398 words · 
Planes parked at gates and crowded passengers at a US airport departure board during the FAA ground stop

FAA Ground Stop: One Router Grounded 20,000 Flights Across the US

The FAA ground stop on Friday, November 13, brought nearly every departing flight in the United States to a standstill for almost five hours. It was the longest nationwide shutdown of US air travel since the September 11 attacks, and federal officials say it may have started with a single piece of networking equipment. More than 20,000 flights have been disrupted since Friday, according to Reuters, which cited government officials. Airlines canceled nearly 10,000 flights on Friday alone and delayed more than 7,000 others, in what became one of the worst travel days in American history. ...

November 15, 2026 · 6 min · 1183 words · 
A laptop showing a code editor with a security warning overlay

Cursor Says Fireworks Breach Led to Fake Claude Code Extension for 5,000 Users

Cursor finally connected the dots on Friday, November 13. The AI code editor published a full incident report explaining how a booby-trapped copy of Anthropic’s Claude Code extension ended up in its marketplace on October 29. The trail starts at Fireworks AI, a vendor Cursor used to train and run models. “An attacker obtained an access key we had issued to Fireworks. The key was exposed by Fireworks,” Cursor wrote in its November 13 update. That one sentence ties together two incidents that looked separate for two weeks: the Fireworks breach Cursor’s users had read about, and the fake extension that showed up in Cursor’s own store. ...

November 14, 2026 · 6 min · 1118 words · 
A crowd of supporters outside a modern glass tech headquarters at dusk

Elon Musk Backs Petition to Reinstate Jensen Huang as Nvidia CEO

The Elon Musk Jensen Huang petition gained its biggest backer on Friday, when Musk said he has signed the open letter calling on Nvidia’s board to reinstate Jensen Huang as chief executive. Musk announced it on X, two days after Huang resigned and ended a 33-year run at the chipmaker. “A petition is underway to reinstate Jensen and I’m signing it too,” Musk wrote on November 13. The post has been viewed more than 8 million times. The letter, on the volunteer-run Stand with Jensen site under the title “Thirty-three years deserve an answer,” had gathered more than 90,000 signatures by Saturday, CNBC-TV18 reported. ...

November 14, 2026 · 5 min · 982 words · 
A security operations center screen showing AI agents scanning application code for vulnerabilities

OpenAI Daybreak puts AI agents on both sides of the cyber fight

OpenAI Daybreak now covers both ends of a security incident. The company launched two new agent products on Thursday, November 12, at its Cyber at the Frontier keynote: Daybreak Red Team, which attacks your software, and Daybreak Defense Factory, which patches it. The pitch is blunt. Attackers already use AI at scale, OpenAI president Greg Brockman told the keynote audience, so defenders need frontier models of their own. “We’re now at a point where companies will be secure only to the extent of the frontier model that they use to defend themselves and how well they utilize that,” Brockman said, according to The Register. ...

November 14, 2026 · 5 min · 918 words · 
A laptop showing an Outlook error screen beside a phone with Microsoft Teams, with a dark office in the background

Microsoft 365 outage ends after 24 hours as Outlook, Teams and Copilot come back online

Microsoft says its Microsoft 365 services are fully back online. The company confirmed the restoration early Saturday, November 14, ending a global outage that knocked out Outlook, Teams and Copilot for more than 24 hours and hit users in at least 74 countries. “We have completed our efforts to address the underlying issue of the incident and restored service to the remaining impacted customers and regions,” Microsoft said in its final status update, posted overnight on its Microsoft 365 Status account. ...

November 14, 2026 · 5 min · 1062 words · 
A massive AI supercomputing campus under construction at dusk with rows of server buildings and power lines in a news illustration

SpaceX Plans $74 Billion AI Data Centers in Texas and Arizona

SpaceX AI data centers are about to get a lot bigger. The company said on Thursday it plans to spend more than $74 billion on two supercomputing campuses, one at its Starbase rocket site in Texas and one in Cochise County, Arizona, in a bet that its AI business is now as important as its rockets. The numbers are hard to ignore. SpaceX expects to spend more than $24 billion at Starbase and roughly $50 billion in Arizona, it said in a statement, Reuters reported. The Wall Street Journal also reported the $74 billion figure. Both projects still need local approval, and both face votes on December 9. ...

November 13, 2026 · 5 min · 982 words · 
A smartphone showing a ChatGPT shopping checkout screen with a payment card and a fingerprint approval prompt

OpenAI launches ChatGPT Wallet so its AI agents can shop and check out for you

OpenAI wants its AI agents to buy things for you. The company launched ChatGPT Wallet on Friday, November 13, a payment feature that lets its dots agents pick products, fill in your shipping details, and complete the checkout — without ever seeing your actual card number. The wallet lives inside dots, the always-on agents OpenAI introduced in September. Ask a dot to find a gift, restock a pantry item, or track down a specific product, and it can now carry the job all the way through to payment. OpenAI is not charging a commission on those transactions at launch, CNBC reported, which makes the play about distribution, not fees. ...

November 13, 2026 · 6 min · 1165 words · 
A smartphone showing a Google Search results page with AI-generated answers

Google Gemini 4 Hydrogen Launches as the New Default for Search AI Mode

Google shipped a new AI model on November 12, and it went straight into the product most people actually use. Gemini 4 Hydrogen is now the default model in Search AI Mode, Google said in its announcement, and it handles the hardest questions in AI Overviews too. The model costs $1.50 per million input tokens, and Google claims it runs 2.5 times faster than the model it replaces. That last number matters more than the name. Hydrogen takes over from Gemini 3.8 Flash as Google’s fast model, and it lands in Search for all users, not just developers or paying subscribers. Free Gemini app users get it through automatic model selection, while developers can call it directly through the Gemini API and Google AI Studio. ...

November 13, 2026 · 6 min · 1082 words ·