Washington Isn't Against AI Safety. It's Against Being Told No.

In April, the Justice Department joined Elon Musk's xAI in suing Colorado over a law that bars AI from discriminating in hiring and lending. Assistant Attorney General Harmeet Dhillon named the state's offense. ^[1]

The Justice Department will not stand on the sidelines while states such as Colorado coerce our nation’s technological innovators into producing harmful products that advance a radical, far-left worldview at odds with the Constitution. — Harmeet Dhillon

Coercion. That was the word for a state telling a company what its AI may not do. The same administration, in the same weeks, was running an actual coercion campaign. It had branded Anthropic a "national security supply-chain risk" — the label normally reserved for foreign adversaries like Huawei — because the company refused to remove the guardrails that stop its models from being used for mass surveillance and autonomous weapons. ^[2][3] The demand had been explicit. Defense Secretary Pete Hegseth said what he wanted from a model. ^[3]

won’t allow you to fight wars — Pete Hegseth

Anthropic refused, and the blacklist followed. But the national-security story never quite held together. The NSA kept using Anthropic's AI even after the sanctions went into effect — the government wanted the technology, just without the restrictions. ^[4] And in April, when the DC Circuit upheld the blacklist, a judge noted the obvious: if the concern were the integrity of the chain of command, the Pentagon could simply stop using Claude. The court ruled instead that "the government's need to secure vital technology during an active military conflict outweighs the potential financial harm to the private company." ^[5] For months, then, the coercion model was legally sanctioned. Only this week did a judge block the designation, ruling that national security was not a blank check to punish and retaliate against government critics. ^[6] But the market had already moved — and it had started moving before any of this. OpenAI removed persuasiveness from its risk framework as early as April 2025. ^[7] In February 2026, as it restructured to take in $41 billion from SoftBank and $13.8 billion from Microsoft, it edited its mission statement. ^[8]

We view this mission as the most important challenge of our time. — OpenAI

The word "safely" was gone. Both moves predate the blacklist. The government didn't create the retreat; it weaponized one that was already underway. After Anthropic was frozen out, the Pentagon fast-tracked startups that don't maintain guardrails at all. Smack Technologies moved its combat planning system forward by more than a year; EdgeRunner AI got a Space Force contract within weeks after waiting a year. ^[9] And the complying company now attacks the resisting one. Sam Altman made the attack public. ^[10]

I guess it’s on brand for Anthropic doublespeak to use a deceptive ad to critique theoretical deceptive ads that aren’t real, but a Super Bowl ad is not where I would expect it. — Sam Altman

Safety researchers were resigning from both labs at the same time — Anthropic's safeguards lead warning the world was in peril, an OpenAI researcher saying the company had stopped asking the questions that drew them there. ^[11] But the target was never safety in general. Google DeepMind keeps publishing and expanding its safety framework — adding manipulation risk categories, releasing a 35-page control roadmap — and faces no blacklist. ^[7][12] The difference is what the guardrails restrict. DeepMind's framework governs how its models behave. Anthropic's guardrails told the government what it could not do — no mass surveillance, no autonomous weapons. That is the line. The military AI order in June made the swap explicit: it prohibits "unlawful" surveillance of Americans — not all surveillance — and replaces company-defined guardrails with state-defined ones. The government now decides what is permissible, not the lab. ^[13] The result is a market where the incentives point one way. Companies without use-case restrictions get expedited contracts; the company that kept them got blacklisted, and its rival now calls its safety stance "elitist." Safety has gone from an industry norm to a competitive vulnerability — and the carrots and sticks are operating at the same time.