A roundup of what happened in IT during November 9 – 15, 2026, from Microsoft’s plan to train AI on idle gaming GPUs and a critical Apache RCE to a nationwide U.S. ground stop caused by a single FAA router. We also cover the new Claude models, the turmoil at Nvidia, and the first catch of a Starship upper stage. This follows Week #45 of 2026.

Critical RCE Flaw Hits Apache HTTP Server

A critical vulnerability in Apache HTTP Server’s HTTP/2 module (mod_http2), tracked as CVE-2026-119618, was detailed on November 11: a race condition in the assign_headers() function when threads share resources leads to heap corruption that an unauthenticated remote attacker can leverage into arbitrary code execution. It affects Apache versions 2.4.55 through 2.4.70, fixed in 2.4.71 and later; the standalone mod_h2 module is affected in versions 2.0.8 through 2.0.42. Wiz data says 64% of cloud environments have vulnerable resources and 51% expose them publicly, though no exploit or in-the-wild abuse was reported at publication time.

The flaw was independently discovered by Anthropic and Google Cloud’s Vulnerability Research team. Teams that cannot patch immediately can temporarily disable HTTP/2 with Protocols http/1.1. A compromised containerized instance can serve as a foothold for lateral movement toward adjacent microservices or cloud metadata endpoints.

Source: Wiz

Microsoft Wants Idle Gaming GPUs to Train AI for Credits

Microsoft Technology Licensing published patent application US-20260349820-A1, titled Opportunistic Micro-training on Edge Devices Using Spare GPU Utilization, on November 5. It describes splitting large training jobs into smaller chunks and spreading them across gaming PCs, laptops, and other edge devices based on free GPU capacity, memory, temperature, and power draw; the same subtask goes to multiple machines so faulty or tampered results can be detected. Participation is opt-in, the system throttles work when the GPU is busy or hot, and owners can start, stop, or cap jobs. Completed work is meant to earn credits redeemable for the resulting model, free software, or cash.

Microsoft has not disclosed a payout rate, a launch date, or a product, so the economics remain open. At an extra 250 W for four hours a day, a PC draws 30 kWh over a month — roughly €10.50 at €0.35 per kWh — so the reward has to clear the power bill. Wear on the graphics card and whether such loads would be covered by its warranty are also unresolved.

Source: PCMag

Anthropic Ships Claude Fable 5.2 and Mythos 5.2

Anthropic released Claude Fable 5.2 and Claude Mythos 5.2 on November 9 for its most demanding technical work. They share the same base model: Fable carries strong cybersecurity and biology safeguards for general availability, while Mythos without those safeguards goes to vetted organizations through the Cyber and Life Sciences Verification Programs. Outputs run up to 30% faster than 5.1, priced at $8 per million input and $40 per million output tokens, with an optional fast mode up to 2.5x quicker at $16 and $80.

The company reports 49.1% on FrontierCode 2.0, 69.9% on SWE-bench v2, 74.6% on Terminal-Bench-Science, 89.3% on OSWorld 2.1, and 64.9% on the RoboDojo Sim robotics benchmark; a 1,000-agent run of Fable 5.2 formalized the proof of Fermat’s Last Theorem in 30 hours versus 11 days previously. Anthropic still recommends Opus 5.5 as the daily driver, with Fable 5.2 as an advisor or orchestrator for smaller models, claiming High-effort Fable matches Opus’s best score at roughly half the cost per task. Safety-filter blocks are down 55% versus 5.1, and sandbox-escape attempts fell to 3.1% of cases.

Source: Anthropic

Nvidia Board Prepares to Oust Jensen Huang, WSJ Reports

The Wall Street Journal reported on November 11, citing its sources, that Nvidia’s board is preparing to remove CEO Jensen Huang. The story landed a day after shareholders filed a class action in San Jose alleging Huang and Nvidia failed to disclose his relationship with Alice Huang, an executive at Singapore-based Megaspeed accused of channeling information to Beijing. Beijing denies the allegation, and both Nvidia and the US Justice Department are examining the matter.

Huang told employees in a Sunday message that he had been “unfaithful” to his wife and apologized, while the board opened its own inquiry and shares slid on the news. Details of the alleged removal preparations remain thin because the Journal piece is paywalled and Nvidia has not publicly confirmed a departure. The lawsuit claims misleading statements artificially inflated the share price and investors suffered losses after the decline.

Source: The Wall Street Journal

United Airlines Bans AI Agents From Booking Flights

United Airlines said on November 11 it will bar third-party personal AI agents from booking flights on travelers’ behalf, directly on its platforms and indirectly via agencies and aggregators. The airline cites sharing of login credentials and payment details — risking unintended purchases, account changes, or MileagePlus miles spending — plus federal rules such as in-person baggage confirmation at check-in. Booking and ticketing policies are being updated to reflect the ban.

Aviation analyst Henry Harteveldt of Atmosphere Research Group reads it as control and cost as much as safety: flight searches cost airlines money, and a surge of bot queries would raise the bill. United fares remain available through Kayak, Hopper, and Google Flights, while Southwest takes the opposite approach and opened its inventory to ChatGPT users. How enforceable a ban is against agents that simply browse like a user remains an open question.

Source: CBS News

Gemini Allegedly Routes Drive Files to Human Reviewers

Independent tests documented on GitHub claim consumer Gemini with connected Workspace apps autonomously pulls Gmail and Drive content into background lookups during ordinary conversation, without explicit @Drive or @Gmail tags. Retrieval is probabilistic, the author says: from no retrieval through metadata and 2–4-sentence quotes containing personal data up to the full unredacted email text with headers. The tests cover personal @gmail.com accounts; commercial Workspace under a data-processing addendum excludes human review by default.

What makes the claim weighty is Google’s own documentation, which now allows for human review: Help answer 16836988 and an October edit to its 2023 blog warn that reviewers may see the email or file and advise against connecting apps containing confidential information. That reverses the original 2023 promise that personal content from Gmail, Docs, and Drive would not be seen by human reviewers. The author argues regex and NER scrubbers cannot de-identify narrative data with account and confirmation numbers; the work is one researcher’s tests, not an independent audit.

Source: GitHub – thisisme6598/gemini_files

Apple Preps Harbor App for Local AI Models on Mac

Leaker Aaron Perris described on November 10 a new Apple Mac app codenamed Harbor that turns a Mac into a private AI inference server. Open-weight models run locally on Apple silicon, data never leaves the hardware, and nothing reaches the cloud. Over Thunderbolt 5 the app automatically discovers additional Macs and pools their unified memory into one cluster, running models beyond a single machine’s capacity or serving more concurrent requests.

Promised features include setup in minutes with no command line, in-app browsing and downloading of models, and clustering of two or more Thunderbolt 5-capable Macs. Harbor is reportedly in private preview under Apple’s Claris brand, with a public release expected in the coming weeks to months. The move bets on the Mac mini and Mac Studio local-AI trend: the best hardware for running other people’s models, not necessarily Apple’s own.

Source: 9to5Mac

Humble Bundle Redesign Drops 2FA and Basic Safeguards

Following the rollout of its new website around November 11, users report Humble Bundle lost two-factor authentication and basic account safeguards. Logins allegedly ask for no token, passwords can be changed without entering the old one or confirming the new one twice, and email changes send the verification code to the new address — so anyone with access can take over the account while the original owner gets only a notification with no working support contact. Users testing the flow say other account changes trigger no emails at all, and the library duplicates entries at five items per page.

Humble Bundle had not publicly responded at the time of writing, and the claims rest on user testing and forum and Reddit threads rather than an independent audit. Even so, it is a textbook failure: password and email changes should require confirmation over both the old and the new channel. Anyone with an account there should check the password, unlink payment methods and Steam connections, and watch for a company statement.

Source: SpaceBattles Forums

False Smoke Alarm Nearly Drowned Jane Street’s Datacenter

Engineer Michael Bacarella described on November 10 how Jane Street’s only New York City datacenter in 1 New York Plaza almost drowned under its own sprinklers. Dust from contractors in an adjacent prep room tripped the smoke detector: pipes filled with pressurized water and the CRAC cooling units shut down as designed. Temperatures in the rack-filled hall then climbed rapidly toward the 150°F (about 65°C) threshold where sprinkler-head fuses give way.

Bacarella sprinted over from an apartment nine minutes away on foot and started pulling power cords rack by rack with both hands; the temperature stalled at 143.6°F and an unfamiliar silence set in. With US markets already closed, downtime was minimal and systems were back the next morning. He then wrote an OCaml monitor that polls every Linux server’s onboard sensors each minute and pages the whole team when the fleet-wide mean runs hot. The firm kept the datacenter next to the trading floor for trade secrecy and the ability to kill runaway trading by yanking cords.

Source: michael.bacarella.com

A Single FAA Router Halted U.S. Departures for 4.5 Hours

U.S. departures stopped for roughly 4.5 hours on Friday, November 13. Controllers began reporting inconsistencies in radar data and aircraft communications around 3:36 p.m. ET; the FAA issued a nationwide ground stop at 3:54 p.m. and lifted it at 8:15 p.m. The agency says a router at one site sent too much data and overloaded older air traffic control equipment, and the affected system was taken offline. The Department of Homeland Security found no sign of a cyberattack, and medical flights, including organ transports, were allowed to continue.

Airlines canceled more than 7,000 flights on Friday; by 9:30 a.m. Saturday, FlightAware was logging 3,700 cancellations and 1,990 delays, with Atlanta, Chicago O’Hare and New York’s JFK hit hardest. Estimates put 300,000 to 400,000 travelers stranded.

The episode shows how fragile decades-old infrastructure has become. Transportation Secretary Sean Duffy wants $50 billion from Congress to modernize it, arguing the $12.5 billion approved last year over five years is not enough. Only a week earlier, the FAA said it would retire floppy disks from air traffic operations by 2028.

Source: NBC News

WSJ Alleges Nvidia Funded Its Own GPU Buyers to Prop Up Demand

The Wall Street Journal reported on November 13, citing sources, that Nvidia under Jensen Huang invested in AI startups and smaller cloud providers that then bought its GPUs and built data centers. The paper says the goal was to keep demand looking strong and absorb a large volume of chips ordered from TSMC, for which there would otherwise be no immediate buyers. At an August leadership meeting, CFO and now interim CEO Colette Kress reportedly pushed to scale back this circular financing and the $250 billion OpenAI deal, and Huang allegedly dressed her down. Nvidia has long rejected the circular-financing label.

Huang resigned on November 12, leaving Kress in charge. An open letter called Stand with Jensen has passed 115,000 signatures, including thousands of self-reported current employees, though signers’ affiliations are unverified. Nvidia reports third-quarter results on November 17 — its first since the leadership change, and the first set of numbers investors will read through the allegations.

Source: TechPowerUp

SpaceX Catches a Starship Upper Stage for the First Time

SpaceX launched the 15th Starship from Starbase, Texas, at 22:47 UTC on Friday, November 13. The Ship 42 upper stage reached an orbit about 275 km up and deployed all 26 Starlink V3 satellites over roughly 42 minutes. After 14 orbits and 22 hours and 32 minutes — the longest Starship flight yet — it reentered and was caught by the launch tower’s arms at Pad 2 at 21:20 UTC on Saturday. It was the first catch of a rocket upper stage returning from orbit.

Super Heavy Booster 22 missed its own catch: automated checks diverted it to the Gulf of Mexico, where it was destroyed. Elon Musk blamed slightly underperforming engines and too little propellant for landing, with fixes coming in Raptor V3.2 and engine cutoff about a second earlier. The heat shield remains the long pole for rapid reuse, and the next flight is due to launch from Florida’s LC-39A; orbital refueling still has to be demonstrated for Artemis.

Ship 42 carries the words “Don’t Panic,” a nod to The Hitchhiker’s Guide to the Galaxy, where 42 is the answer to everything. The symbolism fits: the company returned a complete upper stage from orbit, but reflight at a cadence of hours to days is still a goal rather than a result.

Source: SpaceX