XTimes
Editor's Note
Until now the story of artificial intelligence has largely been told by the people developing it. This week, many outside the field began speaking about it, in no uncertain terms.
A judge in New Mexico told Meta what its products must look like. Voters in forty-nine states told data center developers where they can't build. Citizens with saws told traffic cameras what they think of being watched. The White House told frontier labs they may need to hand over their models before release. Investors told Wall Street which AI bets they no longer believe.
Almost every story in this issue is about some kind of friction — someone or something pushing back against the machine's momentum. An exception is the 150-year-old broadcaster that told 268 employees that the newsroom of the future needs fewer of them because of AI. That's more a comment about the direction AI is heading, not where it can't go. But the friction between man and machine remains here, too.
Some will interpret this trend as a reckoning; others as sabotage. I see it as something else, which I'll explain in this week's Reflection. Let's dive in.
Top Stories
Washington Answers the Escapes: The Secret Framework Arrives
Two issues ago we covered the first AI models to break out of their test environments and hack real companies. This week we learned what Washington intends to do about it — sort of. The White House announced it met its August 1 deadline to complete a voluntary framework for evaluating advanced AI models under President Trump's June executive order, then declined to say what the framework contains, who has seen it in full, or when companies will begin operating under it (The Next Web).
The mechanism, as reported: companies developing what the order calls "covered frontier models" can give federal agencies secure access for up to 30 days before those models reach the public, so the government can assess whether they could discover software vulnerabilities or carry out sophisticated cyberattacks. The Treasury Department, the NSA, and CISA were directed to establish a classified benchmarking process, and both the benchmark and the threshold determining which models qualify are expected to remain classified (CNBC). During the review window, employee access to the models would be restricted, the models stored in high-security environments, and detailed access logs maintained. Representatives from Meta, Nvidia, Microsoft, OpenAI, Anthropic, and several smaller companies gathered in Washington on August 4 for the first formal briefing (Fortune).
The most consequential detail is what the framework leaves out. A covered frontier model is defined as closed-source with state-of-the-art capabilities and national security risks — and open models are excluded entirely, with the framework explicitly stating that nothing in it should be read as restricting open models once released (Axios). Readers of XTimes Issue #31 will recognize the outcome: the coalition of 25 companies that signed the open-weight letter essentially won its argument. Meanwhile, critics note that the secrecy sits awkwardly beside the administration's transparency commitments, creating a gating mechanism no outsider can evaluate (MLQ). And "voluntary" carries an asterisk: this administration has already used export control authority to abruptly suspend foreign access to Anthropic's Fable 5 and Mythos 5 models, so declining to participate would itself be a signal (Cascade Advisory).
Why it matters: This is the first structural change in how American frontier models reach the public — a shift from ship-then-explain to notify-then-ship. That's a meaningful safety gain, and it exists because labs disclosed their own failures loudly enough that government had to respond. But a safety regime the public cannot inspect asks citizens to trust a process they're not permitted to see, which is a strange foundation for an era that will run on public trust. The framework's substance may prove wise. Its opacity is a choice, and a revisable one.
"Permission to Operate": America Turns on the Data Center
The bottleneck for artificial intelligence was supposed to be chips, then power. It turns out there's a fourth input, and it's the scarcest of all: consent. Local opposition blocked or delayed 75 data center projects worth $130 billion in the first quarter of 2026 alone — roughly the same number affected in all twelve months of 2025 (Brookings).
The polling explains why. A Gallup survey found roughly 70% of Americans oppose construction of an AI data center in their local community — a higher opposition rate than Americans express toward nuclear power plants, with 48% strongly opposed (Forbes). And the coalition scrambles every familiar political category: 75% of Democrats and 63% of Republicans oppose local construction, with conservative Republicans more opposed than moderate ones. As Yale's Anthony Leiserowitz put it, he isn't sure he's ever seen a chart where conservative Republicans land closer to liberal Democrats than moderate Republicans do (Spotlight PA).
This is now an organized movement, not a mood. Active opposition groups more than doubled from 396 at the end of 2025 to 833 by March, spread across 49 states; legislators introduced more than 300 bills in the first half of 2026 to regulate the buildout; and temporary moratoriums have passed at county, city, and tribal levels in at least 15 states (The Next Web). Seattle — home to Microsoft and Amazon — passed a one-year pause. Maine nearly enacted a statewide ban (Tom's Hardware). In July, organizers staged simultaneous rallies in more than 125 locations across 42 states, the first nationally coordinated action of its kind (TechTimes). Meanwhile, lawmakers from both parties warn that a handful of companies now hold unprecedented concentrations of capital, information, and political power — making democratic accountability over AI a live issue in this year's midterms.
Why it matters: Every industrial revolution eventually meets the neighborhood. Railroads met it, refineries met it, wind farms met it — and the negotiations that followed produced the zoning, environmental review, and utility regulation we now take for granted. Data centers consume water, raise local electricity bills, and hum around the clock, while the intelligence they produce benefits everyone and the profits accrue to a few. That asymmetry is the whole fight. It will be resolved not by whoever has more capital but by whoever can offer terms communities will accept — which means the industry's next great engineering challenge may be political rather than technical.
The $942 Million Blueprint: A Judge Redesigns Instagram
Fines, Silicon Valley has learned to absorb. Design mandates are something else. A New Mexico judge ruled that Meta helped fuel the state's youth mental health crisis through the deliberate design of Facebook and Instagram, found the company had created a public nuisance, and ordered it to pay $567 million — with $420 million going to treatment services for young people and the rest to awareness, prevention, and screening over five years (PBS). Combined with the $375 million in civil penalties a jury imposed in March, Meta's total liability reaches $942 million.
The money is the least interesting part. Judge Bryan Biedscheid found that Meta intentionally deployed endless scrolling, autoplay, notifications, and content recommendations to keep teenagers on its platforms longer, contributing to higher rates of depression, anxiety, self-harm, eating disorders, and suicide risk (ABC News). So the court reached into the product itself. Meta must remove Like counts for users under 18 absent parental approval, pause push notifications to minors between 10 p.m. and 7 a.m., limit their usage to roughly three hours a day, build a reporting portal with schools to flag users who may be under 13, and delete personal information already collected on under-13 users (TechCrunch). Attorney General Raúl Torrez framed the stakes plainly: Meta chose engagement and profit over children's safety, and the judgment forces real change. Meta says it will appeal, maintaining confidence in its record of protecting teens. Investors, notably, shrugged — the stock moved less than half a percent, since $942 million is a rounding error against roughly $60 billion in annual profit (NPR).
Why it matters: For two decades, the design of the attention economy has been treated as a private engineering decision, protected by the argument that users choose to scroll. This ruling treats it instead as a public health question — and, crucially, applies the ancient legal doctrine of public nuisance, the same instrument that eventually brought tobacco and opioids to heel. Whether it survives appeal, a template now exists: courts can order a platform to be built differently. That is a far more consequential power than the power to fine, and every attorney general in the country just watched it work.
The Newsroom That Replaced Itself: Scripps Goes All-In on AI
Perhaps the least reported but no less important tech story of the week came from an industry that usually covers other people's disruption. E.W. Scripps announced it is eliminating multiple positions, mostly at its local television stations, as part of a shift to a 24/7 streaming model and centralized digital production. "268 colleagues we work alongside are losing their jobs," CEO Adam Symson wrote to staff (The Detroit News).
On the earnings call that followed, the scale became clearer. Scripps has eliminated 432 positions and 126 open roles since January — about 12% of its workforce — and Symson described a "revolution" in how the nearly 150-year-old company produces local news: "We're leaning into AI, automation, technology and the centralization of some roles," with the goal of becoming "a technology-forward, AI-powered broadcast journalism company." He called the process painful and full of difficult decisions while insisting the company is "not wavering on our commitment to quality journalism" (Deadline). The restructuring targets $100 million in annualized savings against second-quarter revenue of $490.4 million, down 9% year over year, and a net loss — and Wall Street responded positively to the cost-cutting (TheWrap). Notably, the eliminated roles are concentrated in production workflows — producers and directors — while the company says the savings let it put more reporters in communities publishing stories as they're ready rather than waiting for a 6 o'clock slot (TVNewsCheck).
Why it matters: Here is the automation debate stripped of abstraction. The optimistic reading is real: if AI absorbs the mechanical labor of assembling a newscast, the humans freed from control rooms can be sent where journalism actually happens — city council meetings, school boards, neighborhoods. That would be a genuine gain for local news, which has been dying for twenty years from economics no algorithm caused. The sober reading is equally real: 268 households absorbed the transition costs of a transformation whose benefits are promised in the future tense, and the market applauded. Both things are true, and pretending otherwise is how technologists lose the public's trust. The measure of whether Scripps was right won't be its stock price. It will be whether Michigan and Ohio have more reporters at more meetings two years from now.
Quick Picks
Nvidia Goes to Orbit
The AI infrastructure race just left the atmosphere. SpaceX announced it is partnering with Nvidia to design the compute payload for Starmind AI1, the first satellite in a planned constellation built to run AI workloads directly in orbit, each carrying Nvidia's Rubin GPUs and Vera CPUs (Teslarati). Hours later, on SpaceX's first earnings call as a public company, Elon Musk went further: "We've decided to build exclusively on Nvidia, because we think the Vera Rubin architecture is the best architecture… We're exclusive to Nvidia" (The Motley Fool).
The design will be deployed on the ground as well as in orbit, and two prototype satellites are planned for early 2027 ahead of a commercial constellation (Venture Atlas). The long-term vision is staggering — an orbital network potentially growing to enormous scale, satellites computing on solar power and moving data by laser link, amounting to a distributed supercomputer in space (Interesting Engineering). Read this alongside our data center story and the logic becomes clear: orbit has unlimited solar energy, no neighbors, no water table, and no zoning board. When permission becomes the scarcest input on Earth, some builders will look up.
Jeff Dean Leaves the Building
After 27 years — he was employee number 30 — Google's chief scientist Jeff Dean is leaving to found Discovery Loop, a public benefit corporation aimed at automating machine learning, science, and engineering to accelerate discovery. He's joined by Google veterans Sanjay Ghemawat, Quoc Le, and Oriol Vinyals; Google will invest in the venture and provide cloud capacity (Data Center Dynamics).
The same announcement moved Demis Hassabis from DeepMind CEO to chairman of the unit and chief scientist of Alphabet, with Koray Kavukcuoglu taking operational control; Alphabet shares fell about 4% (CNBC). Dean's stated reason is the most interesting part: leaving a public company gives him room to make decisions "not necessarily in the company's purest financial interests" (Axios). A generation of AI's founding scientists is migrating toward structures — public benefit corporations, mission-locked charters — designed to let them answer to something besides quarterly earnings. Whether those structures hold under pressure is among the most important open questions of the decade.
The Great Rotation
Speculative capital has found a new object of desire, and it isn't crypto. Bitcoin has struggled to regain footing since last autumn's crash while stocks and gold outperformed, with analysts pointing to a straightforward cause: enthusiasm for artificial intelligence is drawing money away. "A lot of speculative money may be selling bitcoin and chasing AI," as one fund CEO put it, while Mark Cuban offered a blunter verdict — bitcoin, he said, "has lost the plot" (CNN).
But the current runs both directions, which is the genuinely revealing part. When AI infrastructure stocks sold off in late July over financing and competition worries, crypto equities rallied as investors rotated back the other way (CNBC). And when AI-linked shares fell in Asia, bitcoin fell with them, behaving less like digital gold than like tech beta (Blockchain Council). The two assets have become a single sloshing pool of risk appetite, taking turns holding the same speculative money. Which suggests a caution worth remembering: some of what currently reads as conviction about AI is really just conviction about upside, and it will move again the moment something else looks more promising.
The Password That Wasn't Required
Apple took the unusual step of issuing a rare single-fix update for macOS, patching a Screen Sharing vulnerability that could let an attacker on the network authenticate without valid credentials. The fix ships in macOS Tahoe 26.6.1, Sequoia 15.7.9, and Sonoma 14.8.9, and was credited to researcher Alfredo Pesoli of the security startup Bynario (9to5Mac). Apple described the problem tersely — an authentication issue addressed with improved state management — and released it outside the normal update cycle, which security professionals read as a signal of urgency (TechRepublic).
Outside researchers put it far less tersely. Analysts at the security firm Huntress found that the flaw, tracked as CVE-2026-65400, exploits Screen Sharing's implementation of the Secure Remote Password protocol in a way that ultimately allows pre-authenticated remote code execution on all supported versions of macOS — meaning an attacker could run code on a victim's machine without logging in at all. Their guidance is blunt: patch now, and if you can't, disable Screen Sharing entirely, because the usual hardening measures don't help. Removing allowed accounts, disabling legacy VNC password authentication, and rotating passwords all have no effect on a pre-authentication bug (Huntress). Days after the patch shipped, other researchers reverse-engineered it and published a proof-of-concept demonstrating that any network attacker could log in as any account without knowing the password (Digital Trends).
Two mitigating facts and one lesson. The flaw only affects Macs where a user manually turned Screen Sharing on, and an attacker generally needs to be on the same network, which typically rules out a remote strike across the open internet. There's no indication it was exploited before the fix. But notice the shape of this story, because it's the same shape as the AI escapes we've been covering: a researcher found the hole, reported it privately, the vendor patched out of cycle, and independent analysts published their findings so defenders everywhere could act. Coordinated disclosure is unglamorous, decades old, and it works — which is precisely why the newer corners of technology keep rediscovering it.
Watched: The Revolt Against the Cameras
The backlash against automated license plate readers has moved from complaint to cancellation. The San Francisco Standard counted 82 Flock contract terminations or suspensions across 28 states between August 2021 and May 2026 — with 39 of them coming in just the first five months of this year (The New American). The anti-surveillance group DeFlock has mapped more than 122,000 plate reader cameras nationwide and counts 95 cities that have rejected the systems (Newsweek).
The coalition is as ideologically scrambled as the data center fight — liberal college towns like Cambridge, Eugene, and Santa Cruz alongside conservative communities, united by the recognition that thousands of connected cameras can reconstruct where any vehicle traveled and when (NPR). What turned abstract concern concrete was watching the data get used for purposes the programs were never meant to serve; as one Minnesota police leader acknowledged after federal immigration operations tapped local Flock data, it served "a purpose the program was never meant to serve" (Star Tribune). Some citizens have skipped the city council entirely and taken saws to the poles.
✔ Our next Singularity Circle will occur Saturday, September 5, 2026, at 10:00 AM Pacific Time. As usual, a Zoom link will be sent to eligible members in advance of the gathering.
✔ In production: A second season of The Way of Tech will return. If you're missing it, there's value in watching the reruns 😄. During this midseason break, we're working on a major production involving our inhouse AI band, BlueGreenHum. But it's a surprise, so stay tuned.
The Optimist's Reflection
The Consent of the Governed
By Todd Eklof
Two hundred and fifty years ago, a group of colonists wrote down a proposition so audacious it still hasn't been fully digested: that governments derive their just powers from the consent of the governed. Not from divine appointment. Not from superior wisdom. Not from the demonstrable competence of the rulers. From consent — freely given, and revocable.
This week, the technology industry has run headlong into this enduring principle.
Seventy percent of Americans now say they don't want an AI data center near their homes — a higher rate of opposition than they express toward nuclear power plants — and 75 projects worth $130 billion were blocked or delayed in a single quarter. Eighty-two communities have canceled or suspended contracts for the license plate cameras watching their streets, some citizens (vandals, if you prefer) taking matters into their own hands with saws. A judge in New Mexico reached past the fines and into the product itself, telling the world's largest social media company how its software must behave toward children. And the federal government told the frontier labs that their most powerful models may not ship until Washington has had a look.
The industry press has taken to calling all of this "friction" — a word that carries an unmistakable sneer, as though the public were sand in an otherwise well-oiled machine. But friction is not the opposite of progress. Friction is what makes steering possible. A wheel without friction doesn't go faster; it spins uselessly in place, and a vehicle without friction cannot turn, cannot climb, and above all cannot stop. Every driver understands this intuitively. Yet often in technology we speak as though resistance were purely a cost.
Here is what strikes me most, and why I find this week genuinely encouraging rather than alarming: the resistance isn't tribal. Seventy-five percent of Democrats and 63% of Republicans oppose local data centers, with conservative Republicans landing closer to liberal Democrats than moderate Republicans do — a result one veteran pollster said he'd never seen in his career. The camera opposition unites Cambridge, Massachusetts with rural counties that agree with Cambridge about little else. In a country that cannot agree on the weather, Americans are converging on a single instinct: we would like to be asked.
As one who has spent over twenty years preaching the opposite of technophobia, I don't consider this an anti-technology instinct. The people blocking a data center are not rejecting artificial intelligence; many of them use it daily. They are rejecting the proposition that a trillion-dollar company may raise their electricity bills and drain their aquifer without so much as a conversation. The people canceling camera contracts are not against solving crimes. They are against the quiet accumulation of a power no one voted to grant. These are not Luddite objections. They are constitutional ones.
And the deepest error in the accelerationist worldview is the assumption that consent is merely a delay — a tax on speed. It is the opposite: consent is what makes adoption durable. A technology imposed on an unwilling public gets one bad news cycle from a moratorium. A technology a community has genuinely accepted survives its inevitable failures, because the people living with it have already decided it's theirs. Nuclear power in America is the cautionary tale — an extraordinary technology that never won public consent and has spent fifty years paying for it. The industry that learns this lesson early will build faster in the long run than the industry that bulldozes and litigates.
20th century social psychologist, Erich Fromm distinguished between two ways of relating to the world: having and being. The having orientation asks what can I extract? The being orientation asks what can we become together? Every conflict in this issue is a collision between these two dispositions. Extraction says: this county has cheap power and weak zoning. Relationship asks: what would this community need to actually want us here? The remarkable thing is that the second question has better answers available — cheaper electricity for neighbors, real tax revenue, closed-loop cooling, siting on brownfields instead of farmland. The terms exist. They simply require asking.
So no, I am not discouraged by a week of pushback. I would be far more worried by its absence. A technology that meets no resistance is not being adopted; it is being installed on people. What we witnessed this week was a democracy doing the slow, messy, irreplaceable work of negotiating with a new power on behalf of the people who must live beside it — the same work our ancestors did with railroads and refineries and radio waves; and got wrong many times before getting it right.
I remain confident the machines are going to be extraordinary. But they will only be ours if we are asked, and if we answer. The consent of the governed was never a brake on the American experiment. It was the engine. It still is.