↵ select ↓ ↑ navigate esc close

Daily Briefing: November 10, 2026, morning

Daily Briefing ·

Listen: https://serve.buildtall.systems/fabe61930cc79360b7226fd9f757d2e7864a9826f8e1a48cae13886df875f019.mp3


The briefing for Tuesday, November 10, 2026, covers twenty-five articles. The window is dominated by the Nvidia spy affair, a busy week of rocket launches, and a cluster of AI model and tooling releases.

The Nvidia story runs through five articles. The race is on to sue Nvidia, from TechCrunch, reports that Nvidia's stock slid more than four percent after its board opened an investigation into Jensen Huang's relationship with a suspected Chinese spy. One securities law expert says the company's earlier insistence that the relationship was normal could turn into a very expensive problem. A short seller flagged Nvidia CEO's ties to a suspected Chinese spy six months before the scandal, also from TechCrunch, notes that Culper Research was already questioning Jensen Huang's unusually close relationship with a Chinese executive six months before the scandal erupted. Trump Backs Nvidia CEO Jensen Huang Amid Chinese Spy Affair Investigation, from Slashdot, reports that President Donald Trump says the FBI is investigating Jensen Huang's affair with a woman alleged by U.S. intelligence to have gathered information for Beijing. Trump said he spoke to Huang to see how he was doing, and that Huang said he had absolutely no idea the person in question was a spy. Trump described Huang as a very honorable man. Nvidia has launched its own independent investigation into whether Huang was targeted by a foreign intelligence operation. China has denied the allegations, with Foreign Ministry spokesperson Guo Jiakun calling them unsubstantiated smears against China. Jensen Huang reportedly engaged a PR firm with a history of handling high-profile sex scandals a year ago, from TechCrunch, reports that the PR firm was enlisted shortly after an October 2025 New York Times report described Alice Huang's business dealings with Jensen Huang, according to Wired. Jensen Huang's alleged honeypot scandal has inspired some top-notch memes, also from TechCrunch, reports that the internet is having a field day with the alleged honeypot scandal.

A second theme is the week's launch schedule. Launch Preview: Starship, Vulcan, and Electron launches scheduled for busy week, from NASASpaceFlight.com, details seven rockets from four countries set to launch this week. The first launch is a Falcon 9 mission designated USSF-627, delivering a classified payload for the United States Space Force from Vandenberg Space Force Base. The payload quantity is unknown, but some indicators suggest a Starshield launch, the government version of Starlink. Booster B1103 will support the mission on its seventh flight and will land downrange aboard the droneship Of Course I Still Love You. On Tuesday, South Korea is set to launch the second flight of Mir, the military's small satellite launch vehicle, and the first orbital flight of the full-scale rocket. Mir is a four-stage, solid-fuel launch vehicle with a liquid-fueled fourth stage. On Wednesday, a Chang Zheng 2D rocket is scheduled to launch an unknown payload from the Jiuquan Satellite Launch Center. The two-stage rocket can deliver 3,500 kilograms to low-Earth orbit. On Thursday, a Kuaizhou 11 rocket is scheduled to launch from Jiuquan, capable of lofting 1,500 kilograms to low-Earth orbit. On Friday, United Launch Alliance's Vulcan rocket is set to launch from Cape Canaveral on its return-to-flight mission, the first Vulcan flight in over nine months following the ejection of a solid rocket booster on the USSF-87 mission. Vulcan will carry forty Amazon Leo satellites for Amazon's satellite internet constellation, a rival to Starlink. The flight will debut the Vulcan VC6L variant with six solid boosters and a long fairing, and will debut the LEO-optimized Centaur V. Also on Friday, pending regulatory approval, Starship is set to fly its most significant mission yet. Flight 15 will feature the first booster-catch attempt since Flight 8 and the first attempt to catch the ship, roughly twenty-two hours after liftoff. Ship 42 and Booster 22 will both fly for the first time, delivering twenty-six Starlink v3 satellites to low-Earth orbit. Three of the satellites will feature imaging capabilities for imaging Starship's heat shield in orbit. A second Falcon 9 mission, USSF-317, is also set for Friday from Vandenberg, flying another unknown quantity of presumed Starshield satellites. Booster B1100 will support that mission on its eleventh flight. The last scheduled launch of the week is Rocket Lab's Electron from New Zealand on Saturday, named The Mountain God Rises. It will deliver an iQPS satellite for a synthetic aperture radar constellation. The mission will be Electron's one hundredth overall mission, a milestone for the small rocket that first flew in 2017, and the 269th orbital launch attempt worldwide in 2026.

A third theme is AI models and tooling. llm-anthropic 0.31, from Simon Willison's Weblog, reports that Anthropic released Fable 5.2 and Mythos 5.2, and that the llm-anthropic plugin now supports running prompts through Fable 5.2. Willison ran pelican-riding-a-bicycle prompts at all five reasoning levels. The best result came from the xhigh level. The max level failed to return an image at all, running out of tokens after spending 128,000 thinking about them. Willison notes this is the same bug seen with Opus 5.5, but new for the Fable series, since Fable 5 and Fable 5.1 both returned max effort pelicans. Quoting Prompting Claude Fable 5.2, also from Simon Willison's Weblog, quotes Anthropic documentation stating that Claude Fable 5.2 responds closely to what the prompt says about who owns the code it works on. When a prompt said nothing about ownership or presented the code base as someone else's, the model tended to add to the existing structure, even when that repeated existing code. If the prompt says the codebase is the model's own, the model merged repeated code in about half of the tries, compared with under a tenth of the tries without that sentence. Introducing Our Smallest, Most Efficient Model: Muse Ember, again from Simon Willison's Weblog, covers Meta AI's launch of Muse Ember, its entry in the Claude Haiku and GPT Luna category. Meta describes it as its smallest and most efficient model, designed for complex reasoning execution and the work agents repeat all day, including tool calls, coding subtasks, routing, extraction, and long document parsing. For shorter prompts it is priced the same as GPT-6 Luna and Claude Haiku 5.5, at ten cents per million input tokens and fifty cents per million output tokens. Claude Fable 5.2 is eighty times that price. As prompts get longer, pricing thresholds differ across the three models. Willison argues that comparing price per token makes less sense when reasoning tokens mean different models use dramatically different numbers of tokens to solve the same problem. On Artificial Analysis charts, Muse Ember models sit on the Pareto line, the line at which one dimension can only be improved by giving up another. Meta calls this out, saying no model scores higher for less per task. Willison rendered six pelicans riding bicycles at the six reasoning levels. The high level failed to return a full SVG, and the reasoning level did not cleanly map to visual quality. The Ember pelicans are comparable in quality to those from Muse Wonder, which is forty times the price. Anthropic reportedly plans to move forward with IPO plans, from TechCrunch, reports that the AI company is expected to set its valuation range this week, according to the Wall Street Journal.

A fourth theme is AI governance and pacing. Verified Compute Allocation Disclosures as a Foundational Governance Tool, from the AI Futures Project, argues that frontier AI companies should begin pacing as soon as possible, meaning intervening to delay the time at which AIs at a given capability level are developed. The proposed first step is verified compute allocation disclosures, or VCAD. Companies would regularly report how their compute is used, classifying compute used for AI research and development, safety research and development, and external inference. These reports would be verified by third-party auditors, ideally with government support. The authors introduce a framework for thinking about pacing proposals, covering whether pacing is conditional, what safeguards it requires, what restrictions to the inputs of AI research and development it requires, and what needs to be verified to enforce it. They argue that VCAD is crucial to nearly all pacing proposals less extreme than shutting off the GPUs. Their first key claim is that VCAD is required for verifying nearly all restrictions on the inputs to AI research and development, including those that do not directly reference compute, such as delaying internal deployment. The basic reason is that an attack on essentially any input-restrictions pacing regime is to do prohibited things with compute that is not effectively monitored. Their second key claim is that slowing software progress, rather than only delaying training runs, makes pacing much more effective. If only training runs are delayed without slowing software progress, a software overhang results: an undesirable buildup of unused software progress. This could lead to sudden bursts of progress when training runs are eventually approved, and the unused software progress could be stolen by China or diffuse to other actors. The authors describe how VCAD could lead to pacing voluntarily and by mandate. Voluntarily, VCAD supplies trust and public credit, helping resolve a collective action problem in which companies are proceeding faster than they believe optimal. Public disclosures offer a commensurable benchmark for cross-company comparisons. By mandate, most candidate forms of pacing, including caps, floors, cap-and-trade, and conditional requirements, benefit from verified compute allocation. The authors conclude that because pacing is urgent, VCAD is too, and that battle-testing it inside AI companies is urgent.

A fifth theme is health. FDA Warns GLP-1s May Increase Risk of Sudden Blindness, from Slashdot, reports that the Food and Drug Administration said Monday that GLP-1 drugmakers must add a warning label about a rare eye condition that can cause sudden blindness. The condition, nonarteritic anterior ischemic optic neuropathy, or NAION, affects the optic nerve. When blood flow to the nerve is reduced or blocked, it can cause sudden vision loss. The condition is permanent with no known treatment. The FDA said there is a potential increased risk of rare NAION in people taking GLP-1 drugs. The requirement applies to all GLP-1s, including Novo Nordisk's Ozempic and Wegovy and Eli Lilly's Mounjaro and Zepbound, as well as several older GLP-1 diabetes medications. The FDA began looking into a possible link after a 2024 study in JAMA Ophthalmology found diabetes patients prescribed semaglutide were four times as likely to develop NAION as those not prescribed the drug. Cases of NAION have been reported for all currently marketed GLP-1 drugs, but FDA scientists were not able to identify a mechanism. Risk factors for NAION can overlap with conditions seen in people taking GLP-1 drugs, including Type 2 diabetes, obesity, sleep apnea, and hypertension. How A Bedroom Air Filter Could Keep Your Blood Sugar in Check, from Nautilus, reports that new research links clean air to tighter glucose control.

A sixth theme is software and security releases. Microsoft Took 31 Years to Let You Kill Windows 11's Sticky Keys Popup, from Slashdot, reports that Microsoft is testing a Windows 11 option that lets users permanently disable the Sticky Keys and Filter Keys activation prompts directly from the pop-ups themselves, ending a thirty-one-year-old annoyance without requiring a trip into Settings. The dialog is still the same aging Win32 window, but Microsoft's design boss replied under a post saying he is working on modernizing legacy dialogs. The change is in Build 26340.9882 Experimental and Build 26320.9876 Beta, both released on November 6 and based on Windows 11 version 26H2. It is a Controlled Feature Rollout in Experimental, so it may not be visible yet. New Release: Tor Browser 16.0.1, from the Tor Project blog, announces that Tor Browser 16.0.1 is now available. The version includes important security updates to Firefox. The changelog includes updates to Tor, NoScript, and zlib, a rebase onto Firefox 153.7.0esr, and backported security fixes from Firefox 160. Hotel giant Wyndham exposed customer data after app security lapse, from TechCrunch, reports exclusively that Wyndham left a hardcoded credential in its customer app, allowing anyone to access customer data. The hotel giant would not say if it has the ability to check for any past unauthorized access.

A seventh theme is robotics. Boston Dynamics Adds Wheels to Atlas Humanoid, from Slashdot, reports that Boston Dynamics has equipped its Atlas humanoid with wheels, allowing the robot to rapidly roll across flat terrain while still using its articulated legs to navigate obstacles and uneven surfaces. The experiment combined classical model-based motor control with whole-body reinforcement learning to maintain dynamic stability across unstructured, real-world terrain. Atlas also performs pirouettes and various aggressive actuation maneuvers. Andy Wu, controls engineer for Atlas at Boston Dynamics, trained the system in simulation, and the system learned all of the behaviors in sim and executed them in real life with surprising accuracy. A single custom adapter mounts an off-the-shelf scooter rim onto Atlas's existing ankle actuator, and that is the only new part in the whole build. Each wheel is modeled as an ankle that can spin continuously. The next question Wu intends to answer is the policy response to failed actuators. The primary takeaway is understanding control synergy: merging classical motor controls for baseline stability with whole-body reinforcement learning for reactive adaptation yields superior real-world agility compared to using either method alone.

A further theme is California politics. California's proposed 'billionaire tax' fails to pass, from TechCrunch, reports that California voters rejected Proposition 40, the one-time five percent tax on billionaires, according to the Associated Press. Tech leaders outspent backers roughly ten to one to defeat it. Founders and economists interviewed before the vote were split on whether it would have driven out wealthy residents or just trimmed their fortunes.

A final theme is books and film. Unbeatable: The Brutally Honest Case for Free Markets, from Marginal REVOLUTION, announces the new and forthcoming book from Bryan Caplan, published by University of Chicago Press. The excerpt argues that real-world free markets shield humanity from its own Social Desirability Bias, doing good things that sound bad, such as driving incompetent businesses into bankruptcy, producing shameful products people secretly love, and firing superfluous workers so they search for constructive ways to spend their lives. The table of contents includes chapters on what economists really think about markets and government, why economists underrate free markets, why economists overrate government, irrationality as the best case for and against government, and what we ought to do but probably won't. The post says Caplan is very much on target. Godzilla: Minus Zero (no spoilers), also from Marginal REVOLUTION, reviews the film as a morally serious movie, with Ghidorah starring. The review says the whole thing is about twenty minutes too long, but mostly very good. Like many in the Godzilla series, it offers extended and instructive commentary on American-Japanese relations and how the Japanese view them. The reviewer is glad to have gone.

Several articles stand apart from the main themes. Open Thread 459, from Astral Codex Ten, is the weekly visible open thread. It notes a comment of the week from an astronomer saying it is absolutely untrue that there are millions of WOW signals. It also notes that there will be marches in favor of an AI treaty this Saturday, November 14, in major cities across the United States, Canada, and France, including San Francisco and DC. Noah Smith and Azeem Azhar discuss the AI boom, from Noahpinion, describes a conversation between Noah Smith and Azeem Azhar of Exponential View about the AI boom and the possibility of an AI bust. The discussion was sparked by a post Smith wrote a few weeks ago. Smith says they did not reach any firm conclusions, but it was fun to mull the issues over, and he expects more discussions in the future. Locating the source of a mysterious beep with Claude, from Simon Willison's Weblog, describes a project called Beep Locator, a software system for triangulating the source of a mysterious beep using multiple devices scattered around a room. Willison's house has a box room full of unpacked boxes, and something in that room started intermittently beeping. He had Claude Code build the system. Devices play noises to each other to help estimate their locations, and a rough mode allows dragging icons around to provide rough locations. The devices listen for the beep, and a clapping mode helped test the system. The system located the beep to within about forty centimeters, and Willison dug through that pile of boxes and found an ancient carbon monoxide detector that wanted its batteries replaced. The system is based on a 2007 paper called BeepBeep, a high-accuracy acoustic-based system for ranging and localization using commodity devices. Beep Locator runs a Python web application server, and listening devices connect via WebSockets and request browser microphone access. The listeners record a sixty-second buffer of recent audio and run a beep detection algorithm using Fast Fourier Transform and a high-pass filter. When a beep is reported, the server has a device play a short chirp, then asks every listener to upload a clip from before the beep to after the chirp, allowing the clips to be aligned. Willison started the project using Claude Code Cloud via his iPhone with Opus 5.5 in Extra thinking effort mode. He then cloned the repository to his laptop and used Fable 5.2 after it was released. The most interesting part, he says, is that he demonstrated it is possible to give Claude Code ears. Claude does not have native audio input, but it can run JavaScript and Python code that can analyze audio captured by a microphone. Willison could leave Claude alone, hacking away at the problem, querying the audio logs, and reconfiguring its ears. He is reminded of a similar hack at GitHub Universe, where Martin Woodward debugged GitHub Badge computers by plugging them into a computer and pointing a webcam at their screen so a coding agent could tell if updates were working. A Look Back at the Most Breathtaking Photos of Saturn From Voyager 1, from Nautilus, notes that Voyager 1 visited the ringed planet forty-six years ago this month. Daily Reading List – November 10, 2026 (#906), from Richard Seroter's Architecture Musings, is a reading list written from Tokyo. It covers Google's newest TPUs, the 8tx and 8i Max, built for the agentic era; Anthropic's release of Claude Fable 5.2 and Mythos 5.2 with record-high benchmarks and a twenty percent API cost reduction; Meta's release of Muse Ember; building with Google Maps in AI Studio; keeping shipping when GitHub pull requests are slow; AI and the future of the product role; Google Antigravity for knowledge work; a database as a store for global variables; conversational analytics in Gemini with Agentic Data Cloud connectors; developers getting rAIded; a benchmark of coding agent harnesses in which thinking harder beat all of them; managing agents like a team; and voice interactions with coding agents, where Cursor and Google join OpenAI in giving coding agents a voice while Claude Code stays silent.

The window is dominated by the Nvidia affair and by the pace of AI model releases. The Nvidia articles converge on a single unfolding story, while the AI articles range from new models and pricing to governance proposals and hands-on experiments with coding agents.

  1. Open Thread 459
  2. A year after the 100% AI mandate nobody understands our code
  3. Testnet 5 launch incoming - AJ Towns
  4. The race is on to sue Nvidia
  5. Launch Preview: Starship, Vulcan, and Electron launches scheduled for busy week
  6. A short seller flagged Nvidia CEO’s ties to a suspected Chinese spy six months before the scandal
  7. Trump Backs Nvidia CEO Jensen Huang Amid Chinese Spy Affair Investigation
  8. How A Bedroom Air Filter Could Keep Your Blood Sugar in Check
  9. Noah Smith and Azeem Azhar discuss the AI boom
  10. Hotel giant Wyndham exposed customer data after app security lapse
  11. Fable 5.2 one-shotted my nephew's dream game
  12. Jensen Huang reportedly engaged a PR firm with a history of handling high-profile sex scandals a year ago
  13. Locating the source of a mysterious beep with Claude
  14. llm-anthropic 0.31
  15. Microsoft Took 31 Years to Let You Kill Windows 11's Sticky Keys Popup
  16. New Release: Tor Browser 16.0.1
  17. A Look Back at the Most Breathtaking Photos of Saturn From Voyager 1
  18. Horible public key tricks
  19. Locating the source of a mysterious beep
  20. Resurrecting the chess territory
  21. Anthropic reportedly plans to move forward with IPO plans
  22. Verified Compute Allocation Disclosures as a Foundational Governance Tool
  23. CoreWeave bond yield tops 16% as AI credit stress rises
  24. FDA Warns GLP-1s May Increase Risk of Sudden Blindness
  25. Jensen Huang’s alleged honeypot scandal has inspired some top-notch memes
  26. California’s proposed ‘billionaire tax’ fails to pass
  27. Quoting Prompting Claude Fable 5.2
  28. Unbeatable: The Brutally Honest Case for Free Markets
  29. Amazon saves millions by replacing Sonnet 4.6 with Strands Decider
  30. Boston Dynamics Adds Wheels to Atlas Humanoid
  31. Introducing Our Smallest, Most Efficient Model: Muse Ember
  32. Godzilla: Minus Zero (no spoilers)
  33. Daily Reading List – November 10, 2026 (#906)