Washington Rations Access

If You Only Read One Thing

The important shift is not that Washington is regulating tech. It is that Washington is deciding sequence: which countries keep trade access, which customers get frontier models first, and which companies receive strategic exemptions. Read the AP on OpenAI and Anthropic's restricted model access alongside Tariffs Become Platform Policy and The Model Queue Moves to Washington.

Tariffs Become Platform Policy

Digital taxes used to be an accounting fight. They are now a market-access weapon.

President Trump threatened a 100% tariff on imports from any country that imposes a digital services tax on US companies, with Europe as the immediate target. As AP reported, Trump said the tariff would supersede existing trade deals, while the European Commission said it would respond if unilateral measures threatened EU regulatory autonomy. The threat lands just ahead of a July 4 implementation deadline for the US-EU tariff deal that capped most EU exports at 15%, but left digital taxes unresolved.

Why it matters: Digital services taxes are not ordinary corporate income taxes. They tax revenue from activities such as search, marketplaces, social media, and online advertising in countries where the largest platforms may have limited taxable profit. That makes them attractive to governments that think Google, Meta, Amazon, Apple, and Microsoft extract local value without paying enough local tax. It also makes them look discriminatory from Washington, because the affected companies are disproportionately American.

The structural move is that US trade policy is being used as a platform-defense layer. A 2025 White House memorandum already framed foreign digital taxes, content rules, network fees, and data-flow limits as efforts to transfer value from US companies to foreign governments or local champions. USTR has a long Section 301 file on digital services tax investigations, which gives the administration a legal grammar for retaliation. The new part is the scale: a 100% tariff is not calibrated tax diplomacy. It is a warning that taxing the demand aggregator can put your whole goods-export relationship at risk.

This changes the bargaining map for Europe and the UK. A 2% or 3% digital tax raises visible money from a small number of companies; a tariff fight hits exporters, retailers, consumers, and allied diplomacy. That is the point. Trump is forcing governments to choose between taxing US platforms and protecting unrelated exporters that have no direct stake in Big Tech's tax bill.

Room for disagreement: The threat likely starts as bargaining theater. Section 301 takes process, and courts have already constrained some emergency tariff theories. Europe also has domestic reasons to defend the principle that profits should be taxed where users create value. But even a threat that never fully materializes can change incentives. If finance ministries believe a digital tax invites retaliation across autos, wine, machinery, and consumer goods, the expected revenue has to clear a much higher political hurdle.

What to watch: Watch whether USTR opens or renews a formal Section 301 digital-services-tax action. If the tariff threat moves into an administrative record, this becomes more than a social-media negotiating position.

The Model Queue Moves to Washington

OpenAI and Anthropic now have the same product problem: the most important customer is the regulator.

OpenAI released GPT-5.6 Sol in limited preview, but said the first users would be a small group of trusted partners whose participation had been shared with the US government. Its official launch post says the company does not want government access review to become the default, but accepted the step while the administration builds a cyber executive-order process. Hours later, Semafor reported that Commerce Secretary Howard Lutnick allowed Anthropic to restore Claude Mythos 5 access for more than 100 US institutions, while Fable 5 remains unresolved.

Why it matters: The frontier-model market is being reorganized around staged permissions. The old launch model was product-led: announce, publish benchmarks, open API access, and let customers sort themselves by price and capacity. The new model is state-mediated: show the government the model, negotiate the customer list, accept monitoring expectations, and then expand. That shifts release speed from an engineering and safety decision into a political-allocation decision.

OpenAI's details show why this is hard to keep narrow. GPT-5.6 Sol is pitched as stronger at coding, biology, and cybersecurity, with layered safeguards, real-time misuse classifiers, account-level review, and more than 700,000 A100-equivalent GPU hours spent on automated red-teaming. It also says Sol does not cross OpenAI's Cyber Critical threshold. That combination invites a familiar regulatory trap: if the model is not dangerous enough to block, why should Washington choose the first customers; but if it is dangerous enough to choose customers, why should the process be voluntary and ad hoc?

Anthropic's reversal sharpens the same point. Semafor says the new Mythos arrangement removes licensing requirements for named entities and their foreign-national employees, plus Anthropic's own foreign-national employees. That is not just safety review. It is export-control logic moving inside normal enterprise software. Once access depends on named institutions, annexes, citizenship categories, and agency letters, frontier AI begins to look less like cloud software and more like a dual-use industrial base.

Room for disagreement: A phased release can be the responsible answer when models improve at cyber work. AP quoted critics, including Alex Stamos, arguing that blunt restrictions may hurt defenders more than attackers, but the counterargument is serious: cybersecurity capability is dual-use, and mistakes can propagate quickly. The problem is not review. The problem is review without a public, repeatable standard.

What to watch: Watch whether Fable 5 comes back through the same named-institution channel before a general consumer release. If it does, the customer queue will have become the release model, not a one-off exception.

The Contrarian Take

Everyone says: Trump is protecting US tech companies from foreign tax collectors, while the AI limits are about safety.

Here's why that's wrong (or at least incomplete): The common mechanism is allocation power. In the tax fight, Washington is deciding which foreign governments can charge US platforms without risking unrelated exports. In the AI fight, Washington is deciding which customers get scarce model access before the market does. Both moves convert platform power into state-mediated access power. The beneficiary is not simply Big Tech or the regulator; it is whoever can make their commercial priority legible as national interest first.

Under the Radar

Apple's memory workaround has become a China-policy request. After yesterday's price-increase shock, Apple is reportedly seeking US clearance to buy memory chips from China's CXMT, a company on the Pentagon's Chinese military company list. The under-covered point is that AI memory scarcity is now strong enough to push Apple toward a supplier it would rather not explain to Washington. (Source)
Robotaxi regulation is moving from permission to product design. NHTSA proposed removing the manual brake-pedal mandate for vehicles designed to be driven only by automated driving systems. That sounds like a narrow standards update, but it matters because it lets companies design vehicles around autonomous operation from the start instead of retrofitting human controls into machines meant to remove the driver. (Source)

Quick Takes

Hormuz moved from trust failure to enforcement spiral. The US struck Iranian missile, drone, and radar sites after a drone hit a cargo ship in the Strait of Hormuz. Yesterday's issue argued that the ceasefire had not restored operating trust; today's escalation confirms the next test is whether retaliation can reopen shipping or merely make every transit a military signal. (Source)
California's billionaire-tax fight is now a live ballot campaign. Newsom is pushing a national billionaire tax while opposing California's one-time 5% wealth-tax initiative, which is headed to November voters. That resolves the June 25 compromise question: tech wealth did not make the measure disappear, so the next phase is spending, legal prep, and founder-exit signaling. (Source)
Polymarket's integrity problem now includes the web supply chain. Polymarket says a third-party breach let attackers inject malicious code into its site for some users, and that it will refund affected victims. The market-integrity story is no longer only fake creator videos or insider-trading optics; it is whether event markets can secure the front end where users actually place risk. (Source)

The Thread

Today's stories are about who gets to stand at the gate. Europe wants to gate platform profits with digital taxes; Washington wants to gate that gate with tariffs. OpenAI and Anthropic want to ship frontier models; Washington is making the customer list part of the launch. Apple wants cheaper memory; Washington may decide whether China can relieve the shortage. Hormuz shipping wants a ceasefire; navies and insurers are becoming the real arbiters of passage. The old internet story was disintermediation. The new one is reintermediation by states, chokepoints, and scarce inputs.

Predictions

New predictions:

I predict: By August 31, 2026, the Trump administration will publish either a formal covered-frontier-model review process or a named interim customer-approval framework for frontier AI releases, because ad hoc customer-by-customer approvals will become operationally untenable after GPT-5.6 and Mythos. (Confidence: medium; Check by: 2026-08-31)
I predict: Before September 30, 2026, at least one major European government will publicly float delaying, narrowing, or replacing a digital services tax to preserve broader tariff concessions with the US. (Confidence: medium; Check by: 2026-09-30)

Generated: 2026-06-27 03:17 EDT