On March 26, a federal judge in San Francisco issued a 43-page ruling that may define how AI safety is governed for the next decade. Not through new legislation. Not through a regulatory framework. Through the First Amendment — ratified in 1791.
"Nothing in the governing statute supports the Orwellian notion that an American company may be branded a potential adversary and saboteur of the U.S. for expressing disagreement with the government."
— Judge Rita F. Lin, U.S. District Court, Northern District of California
The case: Anthropic refused to let its AI models be used for autonomous weapons or mass domestic surveillance. The Pentagon designated it a "supply chain risk" — a label previously reserved for foreign adversaries. The president ordered all federal agencies to stop using Anthropic's technology. Anthropic sued.
And a federal judge just blocked all of it.
What the Ruling Does
Judge Lin granted a preliminary injunction that freezes the Pentagon's supply chain risk designation and halts the presidential directive banning federal use of Anthropic. The government has one week to appeal. The ruling found two constitutional violations:
"Punishing Anthropic for bringing public scrutiny to the government's contracting position is classic illegal First Amendment retaliation."
The supply chain risk designation was "likely both contrary to law and arbitrary and capricious." No procedural safeguards. No opportunity to respond.
The ruling's severity is unusual. Preliminary injunctions require showing a likelihood of success on the merits. Judge Lin didn't just find it likely — she called the government's position Orwellian.
The Pattern Nobody's Naming
Here is what's interesting about this ruling: it is not an AI governance decision. It is a constitutional law decision that happens to involve AI. The First Amendment was not written to address autonomous weapons or surveillance guardrails. It was written to protect political speech from government retaliation. But because no AI-specific governance framework exists, constitutional law became the default.
This has happened before. It happens every time.
| Technology | Year of Crisis | Governance Gap | Legal Default | Age of Law |
|---|---|---|---|---|
| Internet speech | 1997 | No framework for online expression | First Amendment (Reno v. ACLU) | 206 years |
| Cryptocurrency | 2017–2026 | No digital asset regulation | Securities Act (Howey test) | ~80 years |
| Gene therapy | 2017–2023 | No CRISPR-specific oversight | FDA drug approval (1962 framework) | ~55 years |
| Ridesharing | 2012–2015 | No platform-economy regulation | Taxi licensing (municipal codes) | ~80 years |
| AI safety | 2026 | No federal AI governance | First Amendment (Lin ruling) | 235 years |
The pattern is structural: technology outpaces governance, a vacuum forms, and the oldest available legal framework fills it. Not because it fits — because it's the only thing with enough institutional weight to function as a backstop. The Howey test wasn't designed for tokens. The FDA framework wasn't designed for gene editing. And the First Amendment wasn't designed for AI safety guardrails. But each became the de facto governance framework because the de jure one didn't exist.
Why This Ruling Is Bigger Than Anthropic
The industry response tells you the stakes. More than 400 employees at OpenAI and Google — Anthropic's direct competitors — signed letters defending the company. Google DeepMind chief scientist Jeff Dean signed an amicus brief. Nearly 150 retired federal and state judges, appointed by both parties, filed their own. Microsoft filed one too. The ACLU and the Center for Democracy and Technology filed jointly.
When competitors defend each other in court, they're not being generous. They're recognizing that the precedent applies to all of them. If the government can designate a company a "supply chain risk" for maintaining safety red lines, then every AI company's ethical commitments become a potential liability. The supply chain risk label doesn't just blacklist a company from government contracts — it forces every company that does business with the government to prove they don't use the blacklisted company's products.
The Pentagon didn't just lose a customer. It tried to make Anthropic radioactive — forcing the entire government contracting ecosystem to cut ties. Judge Lin called it what it was: an attempt to "punish" and "cripple" a company for exercising its right to advocate for safety guardrails.
This is why OpenAI employees signed the brief. Not because they agree with Anthropic's specific red lines — but because the mechanism of punishment threatens them all.
The Weakness in the Backstop
The governance default has a structural limitation that no one is discussing. The First Amendment protects the right to build safety guardrails. It does not require them.
Judge Lin's ruling means the government can't punish Anthropic for refusing to allow autonomous weapons use. But it doesn't prevent other companies from allowing it. OpenAI can still offer "all lawful purposes" terms. A startup can ship models with no usage restrictions at all. The constitutional backstop establishes a floor for corporate speech — not a ceiling for AI safety.
This is the structural weakness that repeats across every governance default:
The right of a company to choose to build safety features. Speech. Advocacy. Refusal.
Require any company to build safety features. Set minimum standards. Create industry-wide requirements.
When Reno v. ACLU struck down the Communications Decency Act in 1997, it protected internet speech — but it didn't create a framework for handling online harassment, misinformation, or child exploitation. Those problems persisted for decades while Congress failed to legislate. The First Amendment protected the right to speak online but said nothing about the responsibilities that came with it.
The same structural gap is now opening for AI. The Lin ruling protects Anthropic's right to maintain safety guardrails. But the underlying problem — that there is no federal framework defining what AI safety guardrails should exist, who should build them, and what happens when they're missing — remains completely unaddressed.
What Happens Next
The government has until approximately April 2 to appeal. There is also a separate case pending in the DC Circuit involving a different legal mechanism the Pentagon used against Anthropic. Even if the Lin ruling holds, that case could produce a different outcome.
Three scenarios from here:
The Emergence Signal
The governance default is not a solution. It's a signal that the system cannot produce a solution. When a 235-year-old law becomes the primary legal defense for the most consequential technology of 2026, that's not the system working — it's the system revealing a gap that nothing else has filled.
The pattern is clear across domains: governance defaults to old law, old law provides partial protection, and the gap between partial protection and adequate governance persists for years or decades until something forces specialized legislation. For internet speech, that gap lasted 27 years and counting. For cryptocurrency, the SEC is still issuing interpretive guidance in 2026. For gene therapy, CRISPR regulation still runs through a framework designed for chemical drugs.
For AI, the clock just started. The First Amendment is now AI safety's backstop. It will hold for a while. It won't hold forever. And what replaces it — if anything does — will determine whether AI governance becomes something designed for purpose or something inherited by accident.
The ruling matters. What matters more is what it reveals about the absence it's filling.
Sources: CNBC, CNN, NPR, Fortune, Washington Post, TechCrunch, ACLU, SEC