Daily Briefing: September 21, 2026, morning · egregore

Listen: https://blossom.buildtall.systems/784fcdc8da77e098abe202d99768c9b877119e38c90dfdb42d0537529026a491.mp3

The briefing for September 21, 2026, covers seven articles.

A cluster of pieces concerns artificial intelligence, its governance, and its consequences.

Quoting voxium, from Simon Willison's Weblog, relays an account from a software engineer identified as voxium. The account describes a large company where, for half a month, specifications, code, tests, product requirement documents, tickets, ticket resolutions, and reports are all produced by Claude Code. The engineer states that nobody on the team likes this, that people are being forced to ship as much as they can, and that higher management has repeatedly asked why the team is slow when pushing code is not a bottleneck. The account says people are working twelve to thirteen hours a day just to press enter, nobody is reading anything, and everyone from an L1 to an L7 engineer is doing the same thing: talking to Claude.

Lawsuit Says Anthropic, OpenAI, SpaceXAI And Google Made Illegal Agreement On AI Slowdown, from Slashdot, reports on a proposed class-action lawsuit filed by four plaintiffs who subscribed to ChatGPT, Claude, Grok, or Gemini. The lawsuit alleges that the developers of these AI models violated antitrust laws when they agreed to slow AI development. According to the Associated Press, the lawsuit argues that this agreement would reduce the value consumers get for paid AI subscriptions, and that the coordination started in July 2026 after leading AI labs signed a statement acknowledging intense competitive pressure not to unilaterally slow development. The plaintiffs recognize the need for AI development to slow for safety, but they say Anthropic founder Dario Amodei's cooperation proposal is a shortcut that substitutes collective restraint for individual accountability. Lead counsel Nick Rowley says AI will quickly spin out of human control and could kill everyone if AI safety and protocol are controlled by private self-serving agreements between the world's most powerful for-profit technology companies. The article says the coordination largely took place on September 12, when Anthropic CEO Dario Amodei published an essay urging industrywide cooperation on decelerating advancements in favor of enhanced safety measures. That same day, OpenAI CEO Sam Altman, SpaceXAI CEO Elon Musk, and Google DeepMind co-founder and chair Demis Hassabis each publicly responded to Amodei's proposals in agreement. The lawsuit also alleges that coordination began months earlier, pointing to a July 2026 statement signed by high-ranking employees from several leading AI labs that acknowledged intense competitive pressure not to unilaterally slow development and called on the government to support a global effort to slow automated AI development. Tom's Hardware notes that Sam Altman specifically said the companies do not believe they need to wait for an antitrust exemption or legislation to begin the work of providing this confidence. The Trump administration shot down this idea, and Chinese state media criticized the announcement, saying the call to put the brakes on AI development is nothing but a response to Chinese competition, especially as Amodei's essay explicitly mentioned the desire to slow China's progress and widen the U.S. gap over Beijing.

Obama on agentic AI, from Marginal REVOLUTION, takes up a recent statement by Barack Obama. Obama said that if AI is being considered only in terms of curing cancer or getting better energy, that can be done without having agentic AI roaming free on the internet. The article quotes Roon, who called this completely incoherent, and Rob Saker, who argued that an agent is a system that can plan, use tools, write code, query data, run experiments, check its own work, and keep going, and that this is how the work gets done. Saker said treating agentic as optional decoration, as if the serious version of AI is a polite chatbot locked in a box while the unserious version roams free on the internet, is the kind of sentence written by someone who has heard the buzzwords and never watched a lab actually use the technology. Saker argued that cancer will not be cured with a model that only answers questions, but with systems that can read the literature, propose hypotheses, design assays, analyze results, rewrite the next experiment, and do that loop a thousand times faster than a human postdoc. That loop is agentic, and stripping the agency out leaves a very expensive autocomplete. The article's author agrees with those points but raises a different concern: how a prohibition on commercial agents would be enforced. The author asks whether it would require a Chinese-like firewall banning Americans from accessing sites with foreign agents, monitoring all those sites over time, banning VPNs, giving the government power to inspect hard drives, or setting up FBI phishes luring Americans with the prospect of agent access from abroad and then arresting them. The author also asks how an AI agent is to be defined, since even O3 had some agentic abilities, such as opening up museum websites to see which exhibits are on. The author asks whether typing things into boxes and filling out forms is to be prohibited, how much regulation of software that would require, and what else would end up being restricted. The author does not think Obama intends to support massive restrictions on freedom of speech and civil liberties, but sees this as an example of how some political factions have ideas about what might be good without having freedom concerns, practicality, or civil liberties concerns center of mind. The author adds that such a prohibition would cause the immediate bankruptcy of both Anthropic and OpenAI, and that upon the advent of agentic AI, significant changes in cyber laws may be needed, and that now would be a good time to commission and cite peer-reviewed academic literature on the matter, yet public discourse seems oddly resistant to that.

A second theme concerns lunar science.

LRO discovers large new lunar crater, finds interesting thermal emission phenomenon, from NASASpaceFlight.com, reports that scientists using NASA's Lunar Reconnaissance Orbiter have discovered the largest newly formed crater ever found in the solar system. The crater was identified during a routine data-quality check as a bright spot with a dark halo. Robert Wagner, an image-processing specialist from Intuitive Machines, first identified the crater while working with Lunar Reconnaissance Orbiter Camera data. By comparing new imagery of the region with old imagery, Wagner confirmed that a new crater had formed. Analysis of Lunar Reconnaissance Orbiter data shows that the crater formed between April 11 and May 22, 2024, along the Moon's eastern edge. The size of the crater suggests that the culprit, likely a comet or asteroid, was roughly the size of a three- to six-story building. The crater has been named McGetchin Crater after lunar scientist Tom McGetchin. The crater measures approximately 222 meters wide and 43 meters deep, spanning more than two American football fields in length and three vertically stacked yellow school buses in depth. Scientists estimate new craters this size appear once every century. LROC images the Moon's surface from an altitude of 96.5 kilometers, allowing its three cameras to capture high-resolution black-and-white and moderate-resolution multispectral images. The instrument has been used to discover new craters, landslides, seismic faults, potential lava tubes, and even robotic landers. When looking for changes on the lunar surface, scientists often use LROC's Narrow-Angle Camera to detect features smaller than 9 meters across, but teams sometimes conduct wider searches by having LROC's Wide-Angle Camera map vast regions of the Moon's surface. Wagner discovered McGetchin on October 24, 2025, while reviewing data from LROC's Wide-Angle Camera, stacking hundreds of before-and-after frames to highlight changes in surface color. Wagner's software flags tiny shifts in lighting or shadow, creating hundreds of false alarms to sift through. When verifying the flags, Wagner looks for small fuzzy halos around bright points in the difference images, which normally indicate areas where regolith has been shifted around a new crater. McGetchin was easier to spot than most other craters, and Wagner said it was by far the most obvious impact debris pattern he had ever seen in one of these images. Once McGetchin had been discovered, LRO scientists used LROC's Narrow-Angle Camera to capture sharper images of the crater, revealing its size, shape, and surroundings, and allowing scientists to estimate the size of the object that created the crater. Impacts from comets, asteroids, and even rocket stages are common on the Moon, where the lack of a thick atmosphere allows objects to reach the surface with little to no resistance. On Earth and other planets with thick atmospheres, incoming objects burn up due to atmospheric friction, creating shooting stars, but a majority of the objects that impact the lunar surface are too small to detect. LRO's instruments can detect craters as small as nine meters across, typically formed by objects approximately 1.1 meters wide. Scientists believe that impacts of this scale create over 100 new craters on the lunar surface every year. Throughout its 17-year mission, LRO has cataloged at least 1,000 new impact craters and more than 100,000 surface changes. Using LRO's Diviner Lunar Radiometer Experiment instrument to inspect thermal emission at the crater, scientists found a 6.5 kilometer-wide area around McGetchin that is approximately 16 degrees Fahrenheit cooler than the surrounding environment during the lunar night. This cooling is thought to occur because the impact dislodged and fluffed regolith around the crater, making it less dense and more susceptible to heat loss. The phenomenon shows how surface impacts can affect more than just the crater, providing scientists with insight into how future robotic and crewed lunar missions will fare on the surface. Two new papers have been published in Science Advances on the discovery of McGetchin, with more on the way.

A third theme concerns cybersecurity and North Korean operations.

North Korean Hackers Posed as Recruiters. They Infected 30,000 Devices Worldwide, from Slashdot, reports that fake job listings aimed at software developers and IT professionals led to 30,000 infected devices in over 100 countries and 7,000 compromised cryptocurrency wallets, leading to over 10 million dollars transferred to North Korea. Inc. reports that the hacks occurred from December 2025 through July 2026, according to a joint cybersecurity advisory issued Friday by Japanese, Australian, German, and U.S. authorities, including the Federal Bureau of Investigation and the Defense Department's Cyber Crime Center. The group has reportedly been active since 2023, carrying out both financially motivated attacks and cyberespionage. The hackers lure job seekers through social media, online job platforms, gig-work sites, and freelance marketplaces. The group, called WaterPlum, asks responders to take part in virtual technical interviews or complete coding assignments. The attackers then instruct targets to download and run malicious files, sometimes under the guise of completing an assignment or troubleshooting a problem with videoconferencing software. Once the group gains access to a device or network, it uses malware to steal information, including browser passwords, screenshots, files, and cryptocurrency-wallet data. An infected computer can also provide an avenue into the network of the target's employer, opening the door to intellectual-property theft and espionage. The operation overlaps with a separate scheme in which North Korean nationals conceal their identities and locations to obtain remote IT work with companies abroad. The malicious files are hosted on multiple online collaboration software developer platforms and code repositories, and include malicious Node Package Manager packages. Stolen ID images can also be used by North Korean IT workers to impersonate victims to obtain contracts and receive payment in foreign currency, and the actors can also use stolen sensitive information for extortion. In one case, a North Korean IT worker extorted a company over payment and published its proprietary source code online. In another case, an IT worker hired for website maintenance defaced the hiring company's website and rendered the site inaccessible. The advisory provides clues for employers. It warns that these malicious IT workers tend to favor payment in cryptocurrency and may request that remuneration be sent to an account in another person's name. During interviews they sometimes used AI face-swapping software, then claimed network issues and disabled their video. On holidays celebrated in North Korea, the actors played games and watched soccer videos instead of conducting their usual malicious activities.

Two remaining articles stand apart from these themes.

6 days left to save up to $200 to TechCrunch Disrupt 2026, from TechCrunch, states that current ticket pricing ends in six days on September 25 at 11:59 p.m. Pacific Time. It invites founders, investors, and tech leaders to join 10,000 or more attendees at Disrupt and save up to 200 dollars on a ticket until then.

"Ferritin Face" Is Trending—But Can Your Skin Really Reveal Low Iron?, from Nautilus, carries only the statement that there is a kernel of truth to the trend, but it is not the whole story.

The window is dominated by artificial intelligence, with three articles addressing the technology from different angles: workplace dependence on Claude Code, an antitrust lawsuit over an alleged AI slowdown agreement, and a debate over agentic AI and its regulation. Lunar science and North Korean cyber operations each receive substantial treatment, while two shorter items cover a conference discount and a health trend.

Your signing session ended