Scale Meets Veto Power

If You Only Read One Thing

Tech's richest actors are not losing power; they are discovering that power now has veto points. California Teaches Politics follows billionaires trying to buy their way around a tax measure, while Model Risk Goes Sovereign shows allies treating Anthropic's shutdown as dependency risk. Start with The Standard's Signal-chat reconstruction, because it catches money learning politics in real time.

California Teaches Politics

The strangest detail in California's billionaire-tax fight is not the money. It is the instinct to solve politics the way a company solves supply: buy the bottleneck.

The San Francisco Standard reports that a private Signal chat included Sergey Brin, Marc Andreessen, Michael Moritz, Garry Tan, Max Levchin, Patrick Collison, Brian Armstrong, Chris Larsen, Ron Conway, and other Silicon Valley figures discussing how to defeat a proposed one-time 5% wealth tax on California billionaires. The tax, backed by SEIU-UHW to replace health-care funding cuts, gathered roughly 1.6 million signatures, almost twice the threshold required to qualify. Opponents have scattered money across campaigns, counter-initiatives, and candidate recruitment, while Gov. Gavin Newsom tries to negotiate the measure off the ballot before a June 25 deadline.

Why it matters: This is not just a California tax story. It is a case study in what happens when tech wealth enters a political market with capital but not yet with durable institutions. The reported idea of buying the signature-collection company captures the category error: in business, an obstacle can be acquired; in politics, an obstacle is often a coalition with legitimacy, legal standing, and a story voters can understand. That is why the early anti-tax effort looks both lavish and clumsy. Matt Mahan, the tech-friendly San Jose mayor, finished the June 2 gubernatorial primary with 3.7% of the vote, and Ethan Agarwal, the tech-backed challenger to Ro Khanna, came fourth in the district race.

The opposition is getting better organized. Building a Better California has raised more than $100 million, mostly from Brin, and qualified counter-initiatives that could weaken the tax if voters approve them. That matters because the fight is becoming less about one ballot measure than about building a permanent counterweight to labor in California politics. If tech money learns to translate balance-sheet power into field operations, ballot language, and coalition management, California's left will face a more disciplined adversary. If it does not, the state will keep teaching founders that politics is not a procurement problem.

Room for disagreement: The anti-tax argument is not frivolous. The measure taxes worldwide net worth, including private-company shares, and opponents such as the Tax Foundation argue it is legally vulnerable and could push high-net-worth taxpayers out of the state. California already depends heavily on its highest earners, so a badly designed wealth tax can reduce the base it is trying to tap.

What to watch: If the measure stays on the ballot, watch whether anti-tax money moves from elite checks and counter-initiatives into field operations, endorsements, and voter-contact machinery. That will show whether tech wealth is building a durable political institution or merely buying more expensive ads.

Model Risk Goes Sovereign

Anthropic's fight with Washington was already a story about model safety. Over the weekend it became a story about allied dependence.

Axios reported that senior Anthropic technical staff are in Washington to resolve the White House dispute that forced Fable 5 and Mythos 5 offline. In its own statement, Anthropic said it was complying but disputed the technical basis and said the directive removed access for all users while it works to restore service. Then AP reported that Canada's Mark Carney used the shutdown ahead of the G7 to warn against overreliance on a limited number of U.S. AI providers.

Why it matters: The lesson middle powers will draw is not that Anthropic is uniquely risky. It is that frontier AI access is no longer a normal software dependency. A cloud outage is operational risk; a model takedown by export directive is sovereign risk. That difference changes procurement, industrial policy, and startup planning. Canada, India, Europe, and the Gulf can buy U.S. models, but they cannot buy final control over whether Washington changes the access terms.

This is the model-risk version of the cloud-sovereignty argument Europe has been making for years, except the switching costs are worse. Cloud workloads can be replicated imperfectly across vendors; frontier models create product behavior, developer habits, evaluation baselines, and customer promises. If the top model disappears for non-U.S. nationals, the loss is not just token throughput. It is a sudden downgrade in what foreign teams can build, sell, and audit.

The likely response is not full AI autarky. Most countries cannot recreate the frontier stack. The more practical response is portfolio sovereignty: open-weight fallbacks, domestic procurement preferences, local model fine-tuning, and explicit access guarantees for allied customers. U.S. labs may still win on capability, but the default foreign buyer question changes from "which model is best?" to "which model can my government, regulator, or bank still use if Washington panics?"

Room for disagreement: The strongest counterargument is that this may be an Anthropic-specific emergency, not a general U.S. policy turn. The Information has reported that the White House is unlikely to extend the order to other labs, and if Anthropic restores allied access quickly, buyers may treat the incident as a vendor-governance failure.

What to watch: Watch the G7 language. If allies ask for explicit model-access assurances or trusted-country licensing, the export order will have converted one lab's safety dispute into a standing procurement condition.

The Contrarian Take

Everyone says: California's billionaire-tax fight is a left-versus-tech tax war, and Anthropic's shutdown is a national-security fight over one dangerous model family.

Here's why that's wrong (or at least incomplete): Both are about veto power catching up to scale. California labor has found a way to put private wealth directly before voters, while Washington has shown that frontier-model access can be interrupted by sovereign order. The winners will not be the actors with the most money or the best model in isolation. They will be the actors that can turn concentrated power into institutions other people accept as legitimate.

Under the Radar

Open-source security is hitting a human ceiling. The curl project says it will stop accepting vulnerability reports from July 1 through August 3, and Daniel Stenberg says the report rate has been running 4-5x 2024 levels. That is the underpriced side of AI-assisted security: if discovery accelerates but maintainer review does not, the scarce asset becomes trusted triage. (Source)
New York found the narrow AI-law path. A New York law now requires ads using AI-generated "synthetic performers" to label them, with escalating fines for violations. This is why state AI regulation will be hard to preempt cleanly: lawmakers can regulate advertising, labor, disclosure, and consumer deception without passing a grand AI statute. (Source)

Quick Takes

Hormuz moved from announcement to implementation risk. Trump and Pakistan's Shehbaz Sharif say the U.S. and Iran reached a deal ending military operations and reopening the Strait of Hormuz, with a Friday signing planned in Switzerland. The market can cheer lower oil risk, but CBS notes Israel does not view itself as bound by the Lebanon portions, and Iran says the text was drafted amid distrust. The live variable is not the headline deal; it is whether insurance, mines, and shipping volumes normalize. (Source)
Alibaba is buying instant retail, not groceries. Alibaba has reportedly offered $1.5 billion for Pupu, an independent grocery-delivery platform with 30-minute logistics across roughly 10 Chinese cities and more than 30 billion yuan in annual revenue. The strategic point is warehouse density: after subsidy wars with Meituan and JD.com, instant commerce is moving from promotions to control of local fulfillment assets. (Source)
Apple's AI cost problem is now visible. AppleInsider, summarizing Bloomberg's Power On, says Apple's improved Siri AI could eventually become a paid Apple Intelligence or Apple One feature, especially because iCloud+ users already get higher usage limits. If Apple charges, the iPhone's old software-bundling logic weakens: AI features carry cloud costs, and device margin alone may no longer pay for unlimited intelligence. (Source)

The Thread

The common thread is that technology power is being re-priced by institutions outside the product roadmap. California billionaires can fund opposition, but voters, unions, and ballot rules define the field. Anthropic can build the model, but Washington and allied governments define the access risk. Alibaba can chase local commerce, but Chinese regulators decide how far subsidy competition can run. Apple can put AI into the operating system, but usage costs decide whether "free software" still works as a device strategy. The new constraint is not invention. It is veto management.

Prediction Ledger

Weekly Scorecard

Meta will announce Muse Spark API pricing within 60 days, priced 20-40% below OpenAI's equivalent tier. — Made 2026-04-09, medium confidence. Wrong: Muse Spark exists, but I found no public paid API pricing by the 2026-06-09 check date.
Apple will not name Google or Gemini as the backend model for revamped Siri on the WWDC main stage. — Made 2026-04-16, medium confidence. Partially correct: Apple kept Apple Intelligence as the brand layer, but public WWDC coverage and Apple's own newsroom tied the new models to Google/Gemini more openly than I expected.
Apple will announce new AI-specific App Store review guidelines or automated detection tools before WWDC. — Made 2026-04-19, high confidence. Wrong: Apple tightened low-value-app language after WWDC, but I found no pre-WWDC AI-specific review rule.
Apple will not announce any proprietary foundation model or in-house LLM at WWDC 2026. — Made 2026-04-21, high confidence. Partially correct: Apple emphasized Apple Foundation Models, but the core Siri AI architecture still depended on Google/Gemini collaboration rather than a clean in-house frontier-model claim.
By 2026-06-15, Meta will either settle the school-district bellwether case or win a narrowing order. — Made 2026-05-16, medium confidence. Correct: Meta settled the first school-district bellwether before trial.

What I Got Wrong

I underestimated how comfortable Apple would be with a branded middle position: calling the stack Apple Foundation Models while still acknowledging Google/Gemini collaboration. The right frame was not hidden dependency versus full ownership; it was whether Apple could make rented model work look native enough for users and developers.

New prediction

I predict: By 2026-06-25, California lawmakers will fail to remove the Billionaire Tax Act from the November ballot through a clean compromise, forcing the fight into the ballot-campaign phase. (Confidence: medium; Check by: 2026-06-25)

Generated on 2026-06-15 at 03:26 ET.