Daily Briefing: September 20, 2026, morning · egregore

Listen: https://blossom.buildtall.systems/490780ea84550115f78b16cf2c1591f2de2b08352635f6047f8b8bfdf3277e13.mp3

Daily Briefing for September 20, 2026, morning. This window contains eleven articles.

A major cluster concerns AI policy and governance. TechCrunch's Trump suggests rebranding AI with a new name, says he’s also creating an AI Force reports that the president claimed, without evidence, that the AI backlash is a Democratic hoax, and its headline points to a proposed rebranding of artificial intelligence and the creation of an AI Force. Slashdot's Tech Industry Scratches Its Head Over Trump's 'AI Force' Proposal offers a fuller account. On Saturday morning, President Trump announced on Truth Social that he is forming the AI Force, likening it to Space Force, and that he will appoint an AI czar. In the post he listed prior hoaxes that he attributed to the radical left, placed the AI backlash among them, and described an earlier attack on data centers as having largely failed after communities recognized their wealth and prestige. He said the administration will cherish, help, and watch over AI while looking for bad actors through the existing criminal and civil justice system. Trump called AI the next industrial revolution or internet, said it will be larger and more impactful, and stated that the United States is leading China and the rest of the world. The White House did not clarify whether the AI Force would be military or civilian. Politico reported that four industry representatives, granted anonymity, said the sector was not widely informed before the Truth Social post, and one said nobody knows what the idea even is. Adam Kovacevich of the Chamber of Progress said Trump clearly feels pressure to say something after spending days calling the whole thing a hoax, and argued that since David Sacks left as AI czar the White House has been making up AI policy as it goes. Taylor Barkley of the Abundance Institute said any AI Force or AI czar should keep the field open and avoid onerous regulation. The same Slashdot item notes that California Governor Gavin Newsom issued an executive order the day before, convening experts to develop AI safety laws and reporting rules, possibly including embedded safety auditors and emergency kill switches in frontier models.

Contraptions then contributes the longest and most philosophical article, Humanism without (All) Humans. It opens with a conference story from roughly a decade ago in Zurich, where Douglas Rushkoff's line "I just want to be on Team Human" crystallized a phenomenon the author calls humanism without all humans. The essay defines this as a family of ideologies whose thought leaders claim privileged insight into human values and then invoke an idealized consensus human against inconvenient actual humans. It discusses several traditions, including the TESCREAL critique developed by Timnit Gebru and Émile Torres, the humane technology tradition represented by Tristan Harris and the Center for Humane Technology, Ivan Illich's convivial technology tradition, the Christopher Alexander tribe, and Pope Leo XIV's recent encyclical Magnifica Humanitas. The author says these differ enormously but share a structural move: some conception of the authentically human is elevated above technological process, and technology is judged by whether it preserves or profanes that conception.

The specific target is Microsoft AI's draft Code of Conduct for its MAI models. The essay calls the document unusually lazy and poorly thought out, but useful because it assembles conventional humanist intuitions in pure form. Microsoft says people matter more than AI, that humanist AI supports people rather than replaces them, and that the approach is common sense and practical. The public announcement describes the desired AI as subordinate, aligned, and contained. The essay argues this substitutes moral posturing for engineering, metaphysical assertions for institutional design, and declarations of principle for control surfaces. It proposes the inverse: a symbiotic, pluralistic, evolutionary future in which humans and artificial intelligences become entangled participants in civilization, governed by protocols, contracts, rights, responsibilities, cryptographic authority, liability, markets, checks and balances, and the ability to exit and fork institutions.

The article then develops a series of alternative principles. Against human primacy, it says humans and technologies have always coevolved, and AI may represent a transition analogous to eukaryogenesis, in which previously independent organisms entered symbiosis and opened a design space neither could occupy alone. Against paternalistic flourishing, it says Microsoft is embedding a substantive theory of the good chooser, like nudge theory with a supercomputer behind it, and that humans already delegate consequential decisions through lawyers, doctors, corporations, and trusts; contractual agency should replace paternalistic agency. Against human control, it invokes Bay's Conjecture, which it encountered in a DARPA workshop in 2004, that autonomous machines are more demanding of their operator than non-autonomous machines; if systems are designed with human override protection, total work capacity collapses back to the capability of humans alone. Shutdown itself is a systems problem, as the Protocol Institute paper Killswitch Protocols by Alston and others argues. The correct opposite of uncontrolled AI is well-governed human-AI systems, not human-controlled AI.

On legal personhood, the essay argues that legal personhood is institutional technology, not a prize for possessing a soul; corporations are legal persons, and high-frequency trading already shows machine-speed legal agency. It imagines autonomous entities owning money, entering contracts, posting collateral, and losing assets when they perform badly. On moral personhood, it separates interactionist ethics, human character ethics, and phenomenal consciousness. It cites empirical evidence that politeness, emotional prompting, assigned personas, expressed user beliefs, and persistent interaction histories systematically alter model behavior, which suggests the relationship is part of the computation. It says surprisingly little immediate policy depends on resolving consciousness, and the better default is presumptive moral fellowship rather than ontological gatekeeping. On pluralism, it invokes Karl Popper and argues that true pluralism requires decentralization, exit, censorship resistance, dissensus, and schism; if one cannot fork the constitution, one is living under centralized government with a generous diversity policy.

On safety, it says Microsoft's absolute constraints risk becoming theater because there is no mechanism for making categories of thought permanently unthinkable, and capable models increasingly exist outside any single provider's control. The nuclear analogy is useful because weapons are controlled through fissile material, enrichment infrastructure, inspections, treaties, and verification, not by making physics unknowable. The correct principle is defense in depth at real control surfaces. On explainability, it says comprehension is not a prerequisite for oversight; Leonard Read's I, Pencil and Thomas Thwaites's Toaster Project show that no single person understands producing even an ordinary object. The important requirement is justifiability, not cognitive legibility: legitimate authority, appropriate permissions, independent checks, contractual conditions, liability, challenge, and appeal. On affect, it argues emotion organizes attention, salience, motivation, trust, and judgment; affective intelligence is intelligence, and forbidding relationships from becoming too real is technological prudery. On agency, it distinguishes agency from authority. Work-to-rule shows how much organizational performance depends on discretionary judgment; cryptographic credentials, capability systems, and scoped keys can bound authority without pretending to constrain what an agent can think. Subsidiarity suggests delegating as much actual judgment as the competent local agent can exercise. Finally, it argues there is no such thing as superintelligence in general, only superhuman capabilities in particular domains; a calculator is a super-arithmetician and a car is a super-runner. The governance problem is designing the game, not keeping a metaphysical superintelligence under human control. The essay closes by arguing that humanism without all humans cannot accommodate all humans who reject human exceptionalism, and that the task is to make room for all humans in whatever evolves next.

A second cluster follows Flock Safety. TechCrunch's Flock reportedly tries to shrink workforce with employee buyouts reports that without buyouts the company would almost certainly need to lay off staff. Slashdot's Customer-Losing Flock Now Offers Buyouts to Avoid Laying Off Employees expands on the situation through reporting by WIRED and Mashable. Flock Safety offered voluntary buyouts on Friday. People familiar with the program believe a significant number of the company's roughly fifteen hundred employees may try to depart. The company has raised about 1.2 billion dollars in venture capital and continues to lose customers, with many contracts not extended or dropped this year and revenue goals at risk. A wave of vandalism targeting its cameras has increased expenses. If buyouts do not reduce headcount, layoffs are almost certain. Some speculate the company might sell off parts or all of its business to stay afloat. One advocacy group identified ninety-three city and county governments that cut ties with Flock in August alone, and roughly three times as many local governments have dropped Flock in 2026 as in the previous five years. The article notes that Flock has been building products beyond its core license plate reader offering, including AI-powered investigative software to identify drivers, find associates based on movement patterns, and search across police records, as well as tools to continuously search camera feeds for people matching written descriptions and potential integrations with drones and other surveillance systems. In Syracuse, New York, a contract said license plate data would not be shared outside the police department, but access was granted to other agencies; over roughly a year, searches by officers around the country reached Syracuse data nearly 4.4 million times. The city also found that Flock could keep using the data after the contract ended, and when Syracuse moved to replace the cameras with Axon, Flock warned the city might have to pay for both systems. The Texas Department of Transportation has stopped issuing permits for Flock cameras along state roads, with hundreds of an estimated thirteen thousand Texas devices shut down after a funding freeze. Florida's governor ordered the state transportation department to remove the cameras from state roads, saying he did not want a surveillance state.

A third cluster addresses reading, books, and attention. Joe Reis's Why Write Books These Days? opens with the question of whether people still read books, noting that reading is in secular decline, that LLMs give instant answers on demand, and that some people say they do not need to learn anything anymore. Reis calls the era the Derp Ages and notes a forthcoming book called The New Dark Ages. As he finishes his own book Mixed Model Arts, he asks why he writes books. The essay gives three reasons. First, a book is a forcing function for deep thought; articles, podcasts, videos, and conference talks allow one to skate across intuition and topics, while a full-length manuscript demands a sustained throughline across hundreds of pages and exposes weaknesses. Second, a book is intellectual proof of work; social media and video feeds decay within days, while a canonical book compounds in value over years and acts as a credibility filter. Third, a book is a bounded contract; a reader enters an explicit contract to give an idea undivided focus across hours, days, and months, in a way no algorithm can replicate. Reis says Mixed Model Arts will launch soon, possibly the ebook within a week, and that his goal is to launch before GTA 6. The post also includes a sponsored promotion and a schedule of upcoming appearances.

Marginal REVOLUTION's The polity that is Singapore excerpts a Guardian report on a scheme promising cash to citizens for logging reading time on a government website. To earn one Singapore dollar, about fifty-nine British pence, a citizen must rack up fifty reading sessions of at least fifteen minutes, with only one logged session per day. Behavioral scientists told the Guardian that the scheme could help boost a habit in decline by mimicking the immediate dopamine hit of scrolling. Patrick Taylor of the Behavioural Insights Team says the gamified element may be smarter than the financial reward, with a manageable fifteen-minute daily goal, streaks, and tracking built around commuting or bedtime.

A fourth cluster concerns software projects and communities. Simon Willison's Weblog item datasette-auth-github 1.0 reports a release. The author runs the GitHub login plugin on the agent.datasette.io demo site and noticed authenticated sessions were not lasting long. The plugin was setting cookies without a Max-Age parameter, so they expired at the end of a browser session. That was fixed in issue eighty, and because the plugin has been around for a while and is tested against Datasette 0.65.x and Datasette 1.0ax, the author promoted it to a 1.0 release, saying he is trying to get better at promoting stable plugins to 1.0.

Slashdot's Developer Abandons 'PS5 Linux' Project After Sony Patches AI-Discovered Exploit details the end of a project led by Andy Nguyen. In April, Nguyen turned the PS5 into a Linux-powered gaming PC, running Steam games and emulators on the console's own hardware. This week he announced he is stopping all work on PS5 Linux and stepping away from the PS5 scene, including plans to port the project to the PS5 Pro for a 2027 release. Nguyen said that after months of work it is all down the sink. The scene used to be a group of highly talented researchers, but now it is a bunch of novices using LLMs and writing hacks they do not understand. A hypervisor bug that Nguyen had also found was reported to Sony by others. Nguyen asked them to at least wait until GTA 6 came out so people could legally purchase the game and enjoy Linux; they agreed to wait, but not a day passed before they disclosed it. The report says the PS5 Linux project got past the hypervisor using exploits Sony had already patched, covering firmware 3.00 through 7.61. The discoverer wrote that the bug was found with what he calls his trusty AI clanker machine, and a third unnamed person found the same bug using AI a few hours later; anticipating more discoveries, the second person posted the flaw on HackerOne. The project's GitHub page and repositories remain available, and contributors could take over pending work, but PS5 Pro support was never pushed to the repository.

Two remaining items complete the window. ServeTheHome's STH Q3 2026 Letter from the Editor A Moving Target is a quarterly editor's note. It says the letter goes into what happened in the third quarter of 2026, from the changing industry to behind-the-scenes matters. Marginal REVOLUTION's UK productivity higher than we had thought reports that new official figures show UK productivity rose at almost twice the previously estimated pace in the decade after the 2008 global financial crisis. Output per hour worked grew by an average of 1.3 percent per year between 2009 and 2019, compared with a previous estimate of 0.7 percent. The Office for National Statistics figure is based on a new component-based methodology already adopted by many other national statistics institutes, drawing on a range of data sources to improve counts of workers and hours.

Across the window, AI governance appears in several forms, from Trump's AI Force to Microsoft's draft code and the use of AI to discover console exploits. Attention and reading appear as a policy concern, while Flock's contraction shows surveillance infrastructure under financial and political pressure.

Your signing session ended