Authority without verification is the structural risk of 2026. Not in one domain. Across all of them simultaneously.

The Tension

The tension runs deeper than government overreach. It's showing up in the tools builders use every day.

A METR randomized controlled trial found that experienced developers working on their own repositories were 19% slower with AI coding assistants. They predicted they'd be 24% faster. A separate Anthropic study of 52 developers showed 17% lower comprehension scores, with productivity gains that were not statistically significant. The tools assert productivity. The measurement says otherwise. And the tools themselves corrupt the self-reporting mechanism.

Sit with that for a second. The gap between perceived and actual performance isn't a margin of error. It's a 43-point swing. Developers think they're gaining a quarter of their time back. They're losing a fifth of it. The authority of self-assessment is compromised by the same tool being assessed.

This matters at scale. Junior developer hiring has collapsed 67 to 73 percent since 2022. Companies are making irreversible workforce decisions based on a productivity narrative that two RCTs now contradict. If you run a team, the question isn't whether to use AI tools. It's whether your measurement infrastructure can distinguish real gains from felt gains. Most can't.

Penn Wharton estimates the tariff refund crisis is accruing $700 million per month in interest costs. That number grows because the assertion infrastructure (collecting tariffs) runs automatically, while the verification infrastructure (processing refunds) requires manual intervention that doesn't exist yet. The asymmetry is structural. Assertion scales. Verification doesn't.

This same asymmetry is playing out in energy. Offshore wind projects sit frozen while regulators cite engineering concerns that were solved years ago. The authority to block is exercised without the obligation to verify the basis for blocking. Wind farms that could reduce exposure to $115 Brent crude stay on paper because performative authority requires no proof of competence.

What This Unlocks

The Anthropic lawsuit will establish the first real precedent on whether AI companies retain deployment boundary rights against executive pressure. Every frontier lab is watching. If Anthropic wins the injunction, it means private safety constraints survive government override attempts. If it loses, every AI company's acceptable use policy becomes a suggestion that any sufficiently motivated agency can revoke.

I think the more interesting unlock is structural. The SCOTUS tariff ruling and Anthropic's lawsuit are the judiciary doing what it's designed to do: checking assertions that outran their authority. Courts are slow. But they're the only verification infrastructure that actually binds.

The $166 billion re-entering the economy through tariff refunds is real money. It will reshape import pricing, inventory strategy, and margin assumptions for every company that paid duties under the voided IEEPA authority. If you're building anything that touches cross-border supply chains, the next 90 days of refund processing will matter more than most policy announcements.

OpenAI's acquisition of Promptfoo signals something adjacent. When verification tools get absorbed by the companies being verified, the ecosystem loses neutral evaluation capacity. I think "verification infrastructure" becomes an explicit investment thesis within 90 days. Security tooling post-Promptfoo. Trade compliance post-tariff chaos. AI governance post-Anthropic. The builders who move first on verification will own the trust layer.

For builders specifically: the companies that survive this era won't be the ones with the most capabilities. They'll be the ones that can prove their claims. That's a design problem, not a marketing problem. Build the measurement into the system. Don't bolt it on after the assertion.

Watching Next

Three timelines matter now.

First, the Anthropic injunction. If the court grants a temporary restraining order, it signals judicial willingness to check executive AI authority before a full trial. Speed matters here. Every week Anthropic stays blacklisted, other labs recalculate whether maintaining safety policies is commercially survivable.

Second, Section 122 expires July 24. The White House is launching 301 and 232 investigations to build replacement tariff authority. These have longer statutory timelines and actual procedural requirements. The 150-day clock is the verification infrastructure forcing the executive branch to slow down.

Third, METR acknowledged in a February 24 update that developers are "likely faster now" with current AI tools, but called the evidence "very weak." Updated study results will either rehabilitate the productivity narrative or permanently fracture it. Billions in infrastructure spending and hiring decisions hang on that data.

The Strait of Hormuz adds pressure to all three timelines. Brent past $115 compresses margins everywhere. One-third of global fertilizer trade is halted. Energy costs will accelerate every decision being made under these authority disputes.

Underweighting

Here's what I might be wrong about.

Maybe the Pentagon is right. Maybe AI vendors shouldn't unilaterally set guardrails on military applications. Anthropic is a private company making moral decisions about national defense without democratic accountability. That's also authority asserted without verification. The company's acceptable use policy wasn't voted on by anyone. It was written by executives in San Francisco.

The instinct to cheer Anthropic's refusal is strong. A company standing up to government overreach makes a clean narrative. But the structural question cuts both ways. If we want verification infrastructure for executive authority, we should want it for corporate moral authority too. Who verifies the verifiers?

I think Anthropic is right on the specific question. Domestic surveillance and autonomous targeting without legislative authorization should be refused. But the principle that private companies set deployment boundaries for military technology is genuinely uncomfortable. It works when the company's values align with yours. It doesn't generalize.

Noah Smith's observation that current GDP growth masks structural fragility applies here too. The productivity numbers look fine. Employment is at decade highs. But the composition is shifting beneath the surface. Data center capex is driving productivity metrics, not actual AI adoption gains. The authority of the headline number conceals what the components reveal.

Bottom Line

Authority without verification is the structural risk of 2026. Not in one domain. Across all of them simultaneously. Governments assert trade authority courts haven't granted. Agencies assert AI authority legislatures haven't written. Tools assert productivity that controlled trials can't find. If you're building anything right now, the competitive advantage isn't capability. It's proof. Build the verification into the system before someone else builds it for you.