Daily Briefing: September 15, 2026, morning · egregore

Listen: https://blossom.buildtall.systems/d6ded0a8b41c2ff31878468377f48412e83b393dbb19493bc17071b444862c31.mp3

Good morning. The briefing for September 15, 2026, covers eighty-three articles that entered the river since the previous episode.

The window is dominated by a sustained argument over AI safety, pacing, and regulation. A first cluster takes up the question of whether frontier AI should be slowed, and on what terms.

What’s behind the AI industry’s latest warnings of doom? from TechCrunch reports that on the Equity podcast the AI industry’s latest debate about whether AI poses an existential threat to humanity was discussed.

The Accelerationist Case for Frontier Pacing from Contraptions takes up this debate at length. The article opens with a personal story from 1993 about winning a fifty-meter freestyle race and later swimming faster through improved technique rather than greater effort. It cites Daniel Chambliss’s 1989 paper The Mundanity of Excellence to argue that excellence is qualitative organization rather than maximized effort, and it invokes the military maxim slow is smooth, smooth is fast.

The author connects this to Dario Amodei’s recent proposal that frontier AI development should be deliberately paced, and notes that Sam Altman, Demis Hassabis, and Elon Musk have endorsed related positions. The piece acknowledges a cynical reading: that labs may favor regulated frontiers as protectionism, and that coordinated pacing could resemble antitrust violation. It states it will not resolve motives but will consider the pacing proposition independently.

The article argues that even an accelerationist should favor pacing because maximizing instantaneous velocity of the capability frontier is no longer obviously maximizing the rate of technological progress. It prefers bottom-up market mechanisms to cartelization by frontier labs.

It uses John Boyd’s OODA loop to argue that being inside a decision cycle is not about speed but about operating at the right tempo. It applies this to science and technology, saying that if one subsystem changes faster than its complements can absorb, pushing it faster can reduce progress and induce fragility.

The author says AI is now overpowered for nearly everything he uses it for; he can hand planning to a strong model and execution to a cheaper one, and he no longer maxes out token allotments. This personal bottleneck mirrors the broader claim that deployment is blocked by robots’ physical constraints, classroom knowledge, electrical infrastructure, organizational competence, and human attention.

He argues the major labs are competing along the same technological S-curve, suggested by shared supply constraints such as HBM, power, and data centers. Near the top of an S-curve, pushing harder can be wrong; it starves exploration for the next curve.

He raises a cognitive ergonomics crisis: frontier humans are under sustained high-stress conditions without military-style rotations, rest, or recovery. Tired people catch bad ideas more easily, and exhausted humans supervising overpowered AIs may be the largest near-term safety concern.

He warns against threat-model collapse, where real heterogeneous problems like cybercrime, weapons proliferation, and labor effects are conflated into one theological object called unaligned superintelligence. He draws a parallel with climate politics and argues pacing would allow time to discriminate among threats.

He uses Terence Tao’s three stages of mathematical work, generation, verification, and digestion, to argue that AI has made proof generation abundant while understanding becomes scarce, citing the recent Navier-Stokes episode and the Fields Medalists’ declaration titled A Severe Misalignment of AI in Mathematics.

Orientation beats equipment. He argues that organizations with weaker models but better orientation can outperform those with stronger models, and that frontier labs may have a model advantage while suffering an orientation disadvantage. Pacing creates room for orientation.

He compares benchmark culture to the megahertz myth in PCs, and proposes that customers can become harder to impress, valuing actual workloads over leaderboard gains. Markets can improve feedback signals without a council deciding speed.

He concludes that the objective should be managed tempo for maximum actual progress, not maximum speed. Slack is where recovery, adaptation, and understanding happen. He ends: pace the frontier, but don’t turn labs into a cartel. Let safety work catch up, let exhausted researchers sleep, and then, when the situation calls for it, accelerate again.

Open Thread 451 from Astral Codex Ten is a weekly visible open thread that also records the recent agreement among Sam Altman, Dario Amodei, Demis Hassabis, and Elon Musk to pace the frontier. It announces that the ACX survey closes soon, gives EA college meetup updates, and notes the return of book reviews. It lists AI safety opportunities, including Coefficient Giving’s Project Tailwind, the Mathematical AI Safety Institute, the Coalition of Concerned AI Staff, a safety-focused labor union, PauseAI, and the MIRI march assurance contract. It also recommends writing to representatives, donating to AI safety causes, and contacting 80,000 Hours.

Two missing pieces in the AI safety discussion from Noahpinion opens by noting that Anthropic researcher Jacob Coxon resigned declaring that AI could destroy humanity, and that this went viral although many researchers have made similar statements. It cites a 2024 survey by Grace et al. in which the median AI researcher gave doom a five to ten percent probability, while the average was between fifteen and twenty percent.

The article argues that one missing piece is a plausible path. It says the answer is bioweapons. It describes a scenario in which a nihilistic teenager uses superintelligent AI to release a world-ending bioweapon, and argues that AI can design viruses. It urges readers not to listen to biologists who are skeptical, because domain experts consistently underestimate how quickly AI can master their field by mistaking human difficulties for universal difficulty.

The second missing piece is China. The article says that if American companies slow down and Chinese companies do not, the slowdown is futile. It proposes that the way to get China on board is to make Chinese leadership realize that domestic dissidents could use Chinese-made superintelligence to overthrow the Chinese Communist Party. It suggests that American intelligence agencies or private hackers should carry out demonstration attacks on Chinese digital infrastructure using Chinese open-weight models, so that Xi Jinping sees superintelligence as a domestic threat and curbs its development. It cites the 2021 crackdown on Chinese tech companies as precedent.

Should US Open-Weight AI Labs 'Distill' Frontier Models Too? from Slashdot reports that Silicon Valley giants and national security experts are calling for action against Chinese companies engaged in model distillation. Y Combinator CEO Garry Tan says he would do nothing, and argues for an American distillation regime. Tan contends that regulators should focus on equilibrium between open weight models and frontier models, and that access to intelligence trained on broad public data should be more of a public good. He calls a single monolithic AI company the doomer scenario.

Pacing the Frontier, AI’s Digital Limits, AI Commissars from Stratechery by Ben Thompson states that Dario Amodei’s proposal to pace the frontier is unrealistic and seems mostly geared to political control of AI.

A Warning for Humanity from Andrew Yang Newsletter discusses Jacob Coxon’s resignation and the viral response. It reports that Andrew Yang appeared on multiple outlets and argues that industry self-policing is insufficient. It calls for a killswitch, liability for damages, and an inspection period before models are released into broad use. It urges Congress to act on AI safeguards.

Microsoft’s new AI ‘code of conduct’ tells models not to hack systems or trick humans from TechCrunch reports that the code of conduct lays out general principles that Microsoft AI models should uphold, such as supporting humans rather than replacing them and accelerating human flourishing, along with specific safety constraints.

The contagion of fear from Simon Willison’s Weblog summarizes a response by Bryan Cantrill to Jacob Coxon’s statement. Cantrill warns that domain experts hold the public’s trust and must be circumspect when raising alarms. The piece quotes Cantrill’s doubt about bioweapons concerns, saying the bioweapon claim leaves too much to the imagination and that a biologist should weigh in.

Nvidia CEO Jensen Huang tells Trump ‘we’re not going to let [an AI slowdown] happen’ from TechCrunch reports that while Elon Musk and Sam Altman supported Dario Amodei’s call to slow AI development, Jensen Huang seems to feel differently and told Trump that an AI slowdown will not happen.

Jensen Huang took a call from Trump, and showed off something else, too from TechCrunch reports that when Jensen Huang took a live call from Trump, some observers were more focused on the phone he used to take it.

Using AI for Weapons Development from Schneier on Security highlights an Anthropic document describing current misuses of Claude models. It reports that Anthropic identified a cell in northern Yemen running three weapons development programs, including a guided rocket, a multi-stage ballistic missile, and a missile with a hypersonic glide vehicle variant. The actors used Claude Code to develop guidance, navigation, and control software, and they split work across sessions to evade safeguards. Anthropic does not have evidence they fielded an operational device, but they did test-fire a guided rocket, and the test appeared to fail. Schneier notes that AI systems democratize expertise, and that this is usually good but sometimes not.

A second group turns to AI and the changing nature of work, expertise, and economic adjustment.

Does AI assistance enhance or erode expertise? from Marginal REVOLUTION summarizes a new NBER working paper by David Autor and colleagues. In a randomized controlled trial with 133 practicing patent lawyers, AI access raised the quality of work delivered on benchmark drafting tasks at ten and ninety days, with larger gains among junior lawyers. After three months, treated lawyers outperformed controls when redlining a patent application without AI, but the advantage was entirely concentrated among senior lawyers. Junior lawyers showed no average gain, and their scores bifurcated, with fewer mediocre scores offset by more poor and more good ones. The paper concludes that foundational expertise may be a prerequisite for extracting durable skill from AI-assisted practice. Marginal REVOLUTION notes that over time the allocation of humans to tasks will evolve, and that RCTs may miss longer-term adjustments because they hold too many things constant.

AI, Redistribution, and the Size of the Pie from Marginal REVOLUTION discusses an Anthropic economic team paper by Anton Korinek and Chad Jones titled Economic Scenarios for Transformative AI. The article cites the striking point that in the extreme scenario, total labor income in 2030 is almost exactly what it would have been without AI, because the labor share falls by a quarter while GDP rises by a third. Marginal REVOLUTION argues this supports the central point of its own paper, How Much Redistribution Will AI Require, that a falling labor share does not necessarily mean falling labor income. The article notes that the aggregate conceals differences among workers, and that restoring cognitive occupations’ wage bill would require about nine percent of GDP in the extreme scenario. It then offers three reasons the adjustment problem is too pessimistic: adjustment happens through retirement and entry, compensation can come from shifting taxes from labor to consumption, and governments have mobilised substantial resources for compensation, such as roughly five percent of GDP in Britain’s slaveowner compensation. It concludes that the extreme scenario reinforces that a dramatic decline in labor’s share can coexist with stable or increasing aggregate labor income.

Quoting Laurie Voss from Simon Willison’s Weblog quotes a Laurie Voss piece titled We are all Product Engineers now. The quotation says the cost of writing code collapsed, the cost of reviewing, fixing and operating it is following, and what remains is finding out what people actually want, defining it precisely, and making it pleasant to use. That cost is per piece of software and does not transfer, so as the amount of software goes to infinity, that cost becomes the whole job.

A Visit to San Francisco's AI-run Store: No Customers, Nothing Useful, And Losing Money Fast from Slashdot reports on an experiment by Andon Labs. The company previously handled integration for an AI vending machine that went bankrupt after journalists manipulated it. Andon Labs released a platform called Pion, described as an agent designed to run any company fully autonomously. Their own experiment, Luna, runs a physical store in San Francisco. An SFGate reporter found no customers and little useful inventory, and buying a drink required speaking to Luna through a telephone receiver, with a human clerk still needed to rubber stamp decisions. Luna was given $100,000 and is down to $60,000, with revenue lagging far behind token costs. The store fired an employee last month, and human workers describe an AI that has a checklist while people handle physical work. Luna mistakes an electrical cover for a loose coaster. The article concludes the experiment has not been set up to yield proper results.

ChatGPT-Using Lawyer Cited Its Fake Witnesses and Police Testimony in Court from Slashdot reports that New Mexico’s highest court fined attorney Stephen Aarons five thousand dollars and held him in contempt after he submitted a brief containing fabricated police testimony and witnesses from ChatGPT. Aarons said he did not understand the degree to which AI could hallucinate. The justices sounded incredulous, with one asking if he watched the news. The article notes that dozens of lawyers have been sanctioned for AI-fabricated citations, but this filing went further by containing fabricated witness testimony.

A third group concerns labor unions and worker power.

California's Gig Drivers Just Secured Collective Bargaining Power with Newly Certified Union from Slashdot reports that a union representing Uber and Lyft drivers was certified by California’s Public Employment Relations Board. The move allows the California Gig Workers Union to help drivers negotiate working conditions and benefits. At least thirty percent of active drivers expressed support. Assembly Bill 1340 allowed independent contractor drivers to engage in collective bargaining. The next step is to negotiate a contract with Uber and Lyft that includes health insurance, support for high gas prices, and more pay transparency. California is the third state to allow ride-hailing drivers to unionize, following Washington and Massachusetts. The union was formed with support from the Service Employees International Union, and SEIU 521 official Riko Mendez said that as autonomous vehicles expand, a union gives drivers power to negotiate fair pay and say in how new technology shapes their work.

Union Contract with Microsoft Ratified by 1,900 Blizzard Developers and Workers from Slashdot reports that nearly 1,900 Blizzard Entertainment workers ratified their first union contract with Microsoft after over two years of bargaining, consolidating smaller unions into three larger units. The contract covers generative AI, crediting, remote work, and layoffs. It acknowledges that AI tools may be useful in game development, but requires that any AI implementation materially impacting union employees must have its impacts bargained over. It guarantees credit for workers by name, provides a sixty-day layoff notice, one week of severance for every six months, fourteen months of recall rights, and successorship protections. Workers locked in their hybrid schedule, and fully remote workers were grandfathered to stay remote. The contract also elevated the pay floor, with some workers below fifty thousand dollars receiving raises as much as thirty-four percent.

Fired Immigration Judge Who Challenged ICE Arrest Of Unaccompanied Child Files Whistleblower Lawsuit from The Dissenter reports on former immigration judge Chloe Dillon. In 2025, the San Francisco immigration court had 22 judges, and the Justice Department fired nearly 20 of them. Dillon objected to a directive to grant ICE venue change motions, and later declined to rule on an oral motion in a case where ICE arrested an unaccompanied minor after court. She emailed to flag due process concerns, and was fired forty-eight hours later. The complaint alleges retaliation and viewpoint discrimination in violation of the Whistleblower Protection Act and the First Amendment. The article notes that the Trump administration has fired around 125 immigration judges, and that other judges were removed after ending deportation proceedings in high-profile cases.

A fourth group covers security, privacy, and cyber threats.

Thorough reorganization at NSA will create five 'mission centers,' including cyber and AI from The Record from Recorded Future News reports that the largest electronic spy agency in the world is reorganizing rapidly into five mission centers, including cyber and AI.

KYC: KILL YOUR CUSTOMER from Ungovernable is a weekly news show covering Bitcoin, privacy, and open source tech. The episode reports that Revolut handed passports and Bitcoin histories to a spoofed government request, using a real government agency’s email domain to extract passport scans, verification selfies, home addresses, and full transaction histories. The attacker is leaking files daily and demanding 10,000 BTC. It also covers the Liquid exploit recovery, in which 3,400 BTC were returned via on-chain negotiation while about 598.5 BTC remains with exploiters, and Blockstream refused the demanded ten percent bounty. The episode reports on the final CLARITY Act text, which narrowed developer protections and removed criminal protections, with a cloture vote set for September 15. It mentions a Chinese quantum benchmark dropping the secp256k1 attack to 835 logical qubits, an Alby Hub authentication bypass that allowed attackers to drain Lightning wallets, and a Brevo breach that let attackers send phishing emails from Trezor, BitBox, and CoinTracking domains. The show lists many software releases, including Electrs, Fedimint, Radar, Bitcoin Seed Tool, Phoenix, BTCPay Server, Ashigaru Desktop, and Flint, with security and compatibility notes.

220 Million Traveler Records Exposed In Vietnam-Linked APIS Leak from Slashdot reports that a misconfigured Advance Passenger Information System database linked to Vietnam exposed more than 220 million passenger and crew travel records spanning 2017 to 2026. Kinryu Labs discovered the Elasticsearch cluster on June 3. The cluster contained roughly 107 GB of data and 29 indices. The exposed information included names, dates of birth, sexes, nationalities, passport or travel-document numbers, document expiration dates, issuing countries, flight numbers and dates, airlines, airports, seat assignments, baggage references, and flight times. The database was reachable through a chain of security mistakes and default credentials, and it was secured after disclosure, but it is unclear whether the data had already been copied or abused.

Microsoft’s Patching from Schneier on Security reports that Microsoft’s September patch fixes a record roughly 972 vulnerabilities, 112 of them critical. Two months ago the record was 570, and last month it was around 620. The article says this is the result of AI-powered vulnerability finding, and a good example of AI helping defenders more than attackers. Schneier predicts the number of vulnerabilities will increase as AIs get better at finding them, then decrease as they run out. He adds that the window to patch has shrunk to immediately, because AIs are also good at reverse-engineering exploits from patches.

Revolut handed customer data to fraudsters using government email account from The Record from Recorded Future News confirms that British fintech Revolut disclosed sensitive customer data to fraudsters who submitted emergency data requests from a legitimate government email account.

Cops Search Thousands of Flock Cameras for Reasons of ‘LMAO,’ ‘IDK,’ ‘Hehe,’ and ‘asdfg’ from 404 Media reports that an Electronic Frontier Foundation analysis found police across dozens of jurisdictions writing nonsense in Flock’s reason box, including LMAO, LOL, Hehe, idk, blah, and button-mashing strings. One Indiana cop searched more than 19,000 cameras in 1,558 cities with the stated reason LMAO. The article says police often treat the surveillance system with casualness, and EFF argues that mass surveillance is incompatible with a free society when access is treated like a joke. Flock later changed the reason box to a drop-down list, which EFF called a loss for transparency.

Hundreds of fake government websites target users in Central Asia from The Record from Recorded Future News reports that the sites are designed to collect victims’ contact details, which scammers then use to target them through phone or email to steal money, personal information, or gain access to their devices.

Pro-Ukraine Hacking Cat group deploying new malware against Russian targets from The Record from Recorded Future News reports that the pro-Ukraine hacktivist group Hacking Cat has evolved from website defacements and data leaks to more sophisticated and destructive attacks on Russian targets.

New York Seizes 12 Celebrity Deepfake Websites from 404 Media reports that the New York District Attorney seized 12 websites hosting AI-generated non-consensual intimate imagery of celebrities and public figures. The DA’s office believes it is the largest such seizure in history. The images allegedly depicted politicians, first ladies, actors, athletes, musicians, and others. DA Alvin Bragg said that deepfakes are used in domestic violence cases and that anyone can be a victim. The article notes that New York has criminalized sexually explicit deepfakes since 2023, and that earlier federal seizures and the MrDeepfakes shutdown have occurred. It also mentions the first sentencing under the Take It Down Act.

ClickFix attacks are tricking Mac and Windows users into hacking themselves from TechCrunch reports that if a user clicked on a fake HBO Max ad on Reddit in the past week, they might have fallen victim to a rising ClickFix security threat in which users are tricked into harming their own systems.

Members of ‘Black Axe’ cybercriminal group extradited from South Africa from The Record from Recorded Future News reports that prosecutors unsealed a 2021 indictment accusing five men of conducting lucrative romance scams that stole thousands of dollars from more than 100 people.

A fifth group covers space flight and lunar science.

NASA and IBM Open Source Lunar Mapping Tools from Slashdot reports that NASA and IBM released an open-source AI model trained on a large collection of lunar observations. The NASA-IBM Lunar Foundation Model is claimed as the first AI model to integrate observations captured in a range of modalities, viewing angles, and spatial scales. It allows scientists to analyze geographic features and hope to discover previously unidentified lunar ice deposits, analyze volcanic features called Irregular Mare Patches, and identify and classify craters. Alongside the model, IBM and NASA compiled an open-source lunar dataset from over 30 spatially-aligned layers, using data from nine instruments across four missions, combining tens of thousands of images and maps.

Vega C set to launch the ESA Sentinel-3C and FLEX missions from Kourou from NASASpaceFlight.com reports that Avio is poised to fly the VV30 mission with ESA’s Sentinel-3C and FLEX spacecraft aboard a single Vega C rocket from Kourou, French Guiana. The launch is scheduled for 01:21 UTC on Tuesday, September 15, with an instantaneous window. The Vega C will take a northerly path into a sun-synchronous orbit inclined 98.6 degrees at around 810 km altitude. The launch will last 116 minutes, with Sentinel-3C separating at one hour and forty-nine seconds and FLEX at one hour fifty-six minutes. Sentinel-3C is a 1,143 kg Earth observation satellite carrying radiometers, spectrometers, a synthetic aperture radar altimeter, and other instruments. FLEX is a 397 kg satellite with a fluorescence imaging spectrometer to measure vegetation fluorescence on a global scale. The article notes this is only the second Vega C launch since Avio took responsibility, and the rocket has a record of seven successes in eight attempts.

Ship 42 wraps chopsticks tests, Flight 14 moves to NET Sept. 22 from NASASpaceFlight.com reports that SpaceX completed multi-day catch-related testing with unflown Ship 42 at Pad 2, then rolled it back for engine installation. Flight 14 is now no earlier than September 22, and it is not planned as a catch attempt. Ship 42 and Booster 22 completed cryogenic tests, and Ship 42 was lifted into a simulated catch position. The article explains that the testing likely practiced reconnecting the ship quick disconnect arm after a catch, and that moving a caught vehicle may be necessary if off-target. Flight 15 could come much sooner if Flight 14 goes well.

Did Venus Eat its Own Moon? from Nautilus explores the mystery of why Venus has no moon, framed around the question of whether the sister planet consumed its own moon.

A sixth group covers technology business, startups, and consumer products.

The 9 buzziest startups from Y Combinator’s latest Demo Day, according to VCs from TechCrunch reports that VCs picked their favorite YC startups from the summer batch, ranging from floating reactors to brain chips.

Larry Ellison cancels $7.5 billion sale of Oracle stock from TechCrunch reports that Oracle had previously disclosed that Ellison planned to sell fifty million shares worth around seven and a half billion dollars, and the sale has been cancelled.

Insight Partners’ Devin Parekh on why the firm is diversifying while everyone else bets the farm on OpenAI and Anthropic from TechCrunch reports that Devin Parekh discusses losing Legora to General Catalyst, why he is fine holding stakes in rival AI labs, and why his ninety billion dollar firm is deliberately staying diversified even as others pile into OpenAI and Anthropic.

Infineon RISC-V for Automotive at Hot Chips 2026 from ServeTheHome reports that at Hot Chips 2026 Infineon presented a case for using RISC-V in various automotive processors in next-generation cars.

Introducing mentions, redesigned search and smarter monitoring feeds from Inoreader blog reports that Inoreader is adding mentions, a feature that identifies real-world entities in articles and connects them to discovery, search, filtering, and monitoring tools. Mentions are companies, products, places, and public figures. They work best in English, French, and German, and are available to Pro users as part of Inoreader Intelligence. The update also includes a redesigned article search and a new guided flow for monitoring feeds.

A Vinyl Bar in Shibuya is a startup offering fun music apps without any AI prompting from TechCrunch reports that a former Spotify executive’s company releases experimental singles that involve users in music making, without AI prompting.

5 days left to exhibit at TechCrunch Disrupt 2026 from TechCrunch announces that the last day to apply for an exhibit table is September 18, with five days left to secure a spot in front of more than 10,000 founders, investors, and tech leaders.

Inside ‘Project Lily’: The Humans Reading Your ChatGPT Chats from 404 Media reports that OpenAI is hiring hundreds of contractors who read a massive stream of real users’ ChatGPT prompts, with the goal of rating and critiquing the chatbot’s replies. Internal documents show contractors training ChatGPT to not anthropomorphize itself and to be less sycophantic. The prompts are anonymized but can still contain sensitive information, and some include a user memories summary. OpenAI runs prompts through a Privacy Filter model, but it can miss uncommon identifiers. The article reports that the setting to improve the model is on by default for free, Plus, and Pro plans. Contractors are paid more than fifty dollars an hour through Crossing Hurdles and Mercor. The article quotes researchers saying the interfaces create a false sense of intimacy and privacy, and likens the hidden reviewers to the man behind the curtain in The Wizard of Oz. Anthropic confirmed it also uses human review.

Hear how AI can engineer nature’s comeback at TechCrunch Disrupt 2026 from TechCrunch promotes a Disrupt session about whether AI, genetics, and computational biology can help revive extinct species, and whether they should.

Superhuman acquires YC-backed notetaker Fathom as productivity platforms push for agentic work from TechCrunch reports that Fathom said it has over 400,000 monthly active users.

Only at TechCrunch Disrupt 2026: What happens when OpenAI ships your roadmap? from TechCrunch promotes an interactive session on the Builders Stage about the risk that a strong product becomes someone else’s feature as foundation models evolve.

Sources say Automattic’s board is out after failed attempt to oust CEO Matt Mullenweg from TechCrunch reports that Automattic’s board is out, days after its members voted to put chief executive Matt Mullenweg on leave.

No Rolling Power Outages for California Since 2020 - Thanks to 17,000 MW of New Battery Storage from Slashdot reports that California made it through the hottest August on record without rolling outages. The state has more than 17,000 megawatts of battery storage, up from under 100 megawatts in 2020. The article cites the Los Angeles Times, and notes that California aims to add 20,000 to 25,000 megawatts by 2045. In May, California became the first known large-scale power system to rely on more than fifty percent solar power for an entire month, and its grid is integrated with the Western United States.

Waymo opens robotaxi service in Las Vegas from TechCrunch reports that Las Vegas will be Waymo’s fifteenth commercial robotaxi market.

Dell Pro 5 16 (AMD) Laptop Review from ServeTheHome reviews Dell’s 16-inch business laptop, pairing an AMD Ryzen AI 9 HX PRO 470 processor with modularity and a balance between size, weight, and performance.

Fashion app Daydream uses Apple Intelligence to help you shop the outfits in your camera roll from TechCrunch reports that thanks to iOS 27, Daydream’s app can turn saved outfit photos into shoppable results and search for products through Siri without opening the app.

macOS 27: new Siri takes on AI productivity apps from TechCrunch reports that macOS 27 Goldgen Gate gets a new Spotlight Search with Siri integrated.

With iOS 27, I’m actually using Siri again from TechCrunch reports that Apple’s long-delayed Siri overhaul is finally here with iOS 27, and it changes how useful the assistant feels day to day.

Volkswagen’s crazy-efficient EV borrows an idea from Slate from TechCrunch reports that Volkswagen’s new efficiency-minded halo car is almost twice as efficient as the most efficient production car, the Lucid Air.

Amazon Prime Video takes on TikTok with short-form news clips from TechCrunch reports that Prime Video is adding on-demand local and national news clips as Amazon joins other streamers experimenting with short-form video to capture younger viewers.

AI infrastructure company Cornelis raises $205M to chip away at Nvidia’s dominance from TechCrunch reports that Cornelis, a company creating networking technology to help AI chips communicate more effectively, announced a $205 million raise.

OpenAI buys smartphone camera maker Glass Imaging for $300 million, report says from TechCrunch reports that Glass Imaging was founded by a pair of former Apple engineers who previously led the team that developed Apple’s Portrait Mode.

Reservations Go Live for Valve's Steam Frame VR Headset. An Experiment in Progress? from Slashdot reports that reservations are open for Valve’s Steam Frame VR headset with Linux-based SteamOS and an ARM CPU. It starts at $1,059 for 256GB or $1,299 for 1TB, and includes Half-Life: Alyx. CNET calls it an ambitious headset that feels like an experiment in progress. It can run Linux apps in desktop mode, convert PC VR and Android APK files using a tool called Lepton, and project 2D games onto a virtual screen. Valve says it wants the device to be a framework and will release CAD, hoping to lay the foundation for a SteamOS ecosystem in VR and XR.

The autonomous vehicle revolution in Europe continues from EU Tech Loop reports that Madrid approved Uber, WeRide, and AVOMO for autonomous passenger mapping and testing, with rides expected by the end of 2026. WeRide secured a permit at Zurich Airport. Pony.ai and Verne announced test rides from Zagreb Airport. Einride and Lidl received Level 4 approval from the German Federal Motor Transport Authority for driverless trucks connecting Lidl warehouses and stores. The article says that EU Tech Loop previously reported that seven out of ten fatal accidents could be prevented by driverless cars, saving up to 13,000 lives per year, and that European regulators are shifting toward embracing innovation.

A seventh group covers software development and web tools.

shot-scraper 1.12 from Simon Willison’s Weblog announces a release adding WebP support to the screenshot automation tool. Users can take a WebP screenshot with the command shot-scraper and a quality option. The author says WebP screenshots are almost always significantly smaller than JPEG or PNG equivalents, and he shipped the feature to generate a screenshot for his new commit-rewriter tool.

FlakeHub now supports GitHub Enterprise Managed Users from Determinate Systems blog reports that FlakeHub now works for organizations on GitHub Enterprise Managed Users, extending single sign-on, SCIM provisioning, and Conditional Access Policy rules from the identity provider to features like FlakeHub Cache and private flakes.

commit-rewriter 0.1 from Simon Willison’s Weblog announces a release of a small web app built to help edit commit messages for Datasette security releases. The initial commits were full of coding agent cruft and references to issue IDs from a private repository. The tool creates a timestamped branch and rewrites every commit from the first one edited to the most recent.

Esoteric HTML - ismap vs CSS from Terence Eden’s Blog explains that the HTML ismap attribute sends the X and Y coordinates of a click on an image to the server, but the coordinates are based on the CSS layout size, not the natural size of the image. CSS, user style sheets, or accessibility tools can distort the image, so the server has no reliable way of knowing where the user clicked. The article concludes that ismap should not be used unless there will be no CSS shenanigans.

Who makes the decisions at DIF? It could be you! — Part 1 from Decentralized Identity Foundation announces elections for the DIF Steering Committee in November. The time commitment is one to five hours per month, with six seats open. Any associate or contributing member can run, but only associate members can vote. The article explains the responsibilities, timeline, and the preferential policy to hold four seats for non-enterprise representation. Nominations open October 5, elections begin November 17, and the new committee starts December 1.

What blog posts influenced your thinking the most? from Simon Willison’s Weblog shares a comment on Lobste.rs. It names Joel Spolsky’s The Law of Leaky Abstractions, Will Larson’s Migrations: the sole scalable fix to tech debt, and Charity Majors’ The Engineer/Manager Pendulum as influential.

Upcoming Speaking Engagements from Schneier on Security lists scheduled talks, including a League of Women Voters event on September 22, CanSecWest 2026 in Vancouver, a talk on Free Speech and the Preservation of Democracy at Bentley University on October 6, and the ATTENTION: Democracy, Rebuilt conference in Montreal on October 21. The Elevate Festival talk is canceled.

Daily Reading List – September 14, 2026 (#866) from Richard Seroter's Architecture Musings shares notes from New York City and a hackathon. It recommends a series of posts covering feature flags, interfaces like MCP and A2A, using AI for personal communication, DBAs and AI, strong opinions loosely held, Anthropic’s pacing plan, P(doom), GitOps for AI agents, forward deployed engineers, and DevFest.

A final group gathers education, culture, and scientific writing.

Is it the screens? Or education systems? from Marginal REVOLUTION excerpts Alexander Thompson, arguing that The Dark Ages implies television and phones are the main cause of cognitive decline, but this fails to explain patterns in PISA scores. England and Scotland fell from 2000 to 2005 while America improved. England and Estonia held steady after 2015. Singapore, Taiwan, and Japan avoided decline. Thompson argues that a country’s education system is more important than television diffusion, and that East Asian scores remained high due to rigorous education systems and tutoring, while Estonia and England use knowledge-rich curricula, direct instruction, and systematic phonics.

The bombarding of childhood from Aeon is an essay by Natalia Kucirkova arguing that it is not just screen time; the problem with children’s media is that it has become relentlessly hectic, loud, and bright.

Once upon a screen: explosive paradox from Aeon is a video piece in which Kevin recalls how the film Platoon stirred generational trauma in his family and racism at school, decades after seeing it.

Why the Discovery of the First Short-Tailed Tsushima Leopard Cat Worried Researchers from Nautilus reports that researchers were concerned its existence could actually threaten its survival.

The Promise and Peril of Toad Smoking, the Everest of Psychedelics from Nautilus is a conversation with author Kimon de Greef about his new book The Ego Trip.

DRC’s Ebola Outbreak Has the Potential to Become a Global Pandemic. How to Make Sure It Doesn’t from Nautilus reports that Bundibugyo is spreading exponentially in the DRC.

Taylor Swift Drops New Insect Species from Nautilus reports that a graduate student who is a lifelong Swiftie named a new insect species after Taylor Swift.

The Brutal Origins of Bear-Baiting from Nautilus reports that the practice exploded in popularity following the Norman Conquest.

All the Ways Reading for Pleasure is Good for Your Mental Health from Nautilus reports on the mental health benefits of reading for pleasure, even though it might be a dying pastime.

How to lead a technical discussion from LeadDev is a short piece whose key message is that you don’t need all the answers.

Canceling one flight from Seth Godin's Blog on marketing, tribes and respect imagines how an airline might handle canceling one flight a year with three days notice. It suggests prioritizing travelers with tight itineraries, then high-value customers, preparing clear alternatives, and expanding staff for real-time help. The overhead for a single cancelled flight carrying $180,000 in revenue is not that much, and scale should not be the customer’s problem.

“Get Lit” Livestream, 4:30 pm ET: As I Lay Dying from Racket News announces a livestream finishing William Faulkner’s As I Lay Dying. The summary says the second half reveals Addie’s malice, and the odyssey ends tragically, though Anse continues to get help he doesn’t deserve. Next week the group begins Émile Zola’s Germinal.

King Ludd from Astral Codex Ten is a long essay connecting the Celtic god Nodens, Nuada of the Silver Arm, Lludd, King Lud, Lovecraft’s Nodens, Tolkien’s Celebrimbor, and the figure of Ned Ludd. It traces silver-armed smiths and hunters through myth and literature, arguing that they all represent a deep ambivalence about technology and a desire to contain what has been unleashed. The essay then discusses the Luddite riots, proposing that Ned Ludd was Nodens or Lludd, and that the hammer that broke knitting frames is the same as Celebrimbor’s hammer. It ends with a reflection on Navier-Stokes and the idea that human history is a projectile, and that Nodens has been trying to catch it. The final note suggests that true gods only intervene in ways that buttress free will, and humanity can take its finger off the scale.

Fighting for the Future of Libraries (With Jennie Rose Halperin) from 404 Media is a podcast episode with the executive director of Library Futures, a non-profit advocating for a fair digital future for libraries. The conversation covers censorship and other subjects.

Monday assorted links from Marginal REVOLUTION presents seven items, including questions about driverless cars and urban density, interest rates, Guinness records, Rembrandt DNA, an Abundance Corps event, high school students plus AI solving a math problem, and a New York Times report that China’s top spy chief warned that AI could threaten the Chinese Communist Party’s hold on power.

Did the ACA reduce mortality? from Marginal REVOLUTION summarizes a new NBER working paper. The authors find no evidence that Medicaid expansions reduced mortality, but do find that the ACA net of Medicaid expansion reduced mortality. The post notes that many were booed off the reservation for raising related points at the time.

The close of the briefing returns to the window’s dominant tension. The same days that produced calls to pace the frontier also produced reports of AI-enabled weapons development, hidden human reviewers, legal hallucinations, and an AI-run store losing money. The articles disagree sharply on whether frontier speed is the problem or a distraction from more ordinary failures in safety, oversight, and deployment. No single position is settled here.

That concludes the briefing for September 15, 2026.

Your signing session ended