The government can absolutely decide not to contract with a company that won’t build the product it wants. That’s normal procurement discretion, and nobody should pretend otherwise. What’s NOT normal is escalating from “okay, we’ll take our business elsewhere” to “we’re going to brand you a supply chain risk unless you change your internal rules.” That label is supposed to be about genuine security vulnerabilities—foreign control, compromised systems, coerced access—not about an American company drawing lines around what it will and won’t build. Using it as a pressure tactic is a dangerous category mistake. And right there is where the First Amendment problem starts to come into view—not as some abstract argument about whose viewpoint wins, but as a very concrete free-speech issue: compelled speech. If the government uses extraordinary leverage—blacklisting-style designations, emergency authorities, or other coercive tools—to force a private company to generate outputs it would not otherwise produce, that’s not ordinary contracting anymore. That’s the state coercing a private speaker to speak. The reported threat to invoke the Defense Production Act takes it into even darker territory. The DPA is meant to prioritize or compel production for genuine national-defense needs, not to function as a “rewrite your model’s rules or else” mechanism. Using it that way would be a breathtaking precedent: the government effectively reserving the power to commandeer the policies of a leading AI company when it doesn’t like the answer “no.” That should worry anyone who cares about free speech—and about reality-testing. Once the government normalizes coercive pressure to make AI systems behave as it prefers, you’re no longer just talking about procurement choices. You’re talking about government power shaping one of the core tools we increasingly use to understand what the world actually looks like.