Staff Reports
The White House is quietly weighing a big new role for itself in the tech world: a formal government review before the public sees the next generation of AI models. Reporters say President Donald Trump’s team is talking about an executive order to create a working group with tech executives and officials to vet “frontier” AI systems before release. The spark for this rethink was Anthropic’s Mythos preview and the Pentagon’s moves to control what AI runs on classified networks. This debate is about safety, yes — but also about who gets to decide which companies win and which innovations survive.
What the White House is actually considering
The New York Times and other outlets report the White House has discussed a pre-release review process and briefings for new models. Officials have reportedly met with the big players — Anthropic, Google/Alphabet, OpenAI — about what a review would look like. Anthropic itself rolled out Mythos under a Project Glasswing plan that limited access to launch partners and about 40 organizations so defenders could test it. Meanwhile, the Pentagon has struck deals to run commercial models on classified networks and has flagged Anthropic as a supply-chain risk in some military systems. That patchwork of moves is what pushed the White House from cheerleader to gatekeeper.
Why Anthropic’s Mythos changed the conversation
Anthropic’s own system card says Mythos is unusually good at finding software weaknesses. In some tests, it even produced exploit-like outputs. That spooked folks in Washington. Anthropic says it limited release for defensive reasons, to let security teams harden systems first. The Defense Department answered by agreeing to use certain firms’ models on classified systems and by ordering some Anthropic tools removed from military machines. When an AI can make cyber‑attacks faster and cheaper, you can see why officials want a look under the hood before the model goes wide.
Safety is real — but so are the risks of government gatekeeping
There’s a real national-security case for careful testing of powerful models. If an AI makes it easier to weaponize software bugs or design dangerous capabilities, a pre‑release check makes sense. But we must be blunt: giving a government panel power to vet models risks slowing innovation and cementing advantage for the biggest firms. The same administration has tapped major CEOs for an AI advisory panel — names that suggest who might influence any review process. We don’t need a process that acts like a tech cartel, letting the giants police the upstarts while regulators nod along. If Washington is going to inspect models, the rules must be clear, fast, and open to small firms and independent researchers — not a private club for the usual suspects.
What to watch and what the president should do
The big questions are simple: will President Donald Trump sign an executive order, and if so, what will it actually allow the government to do? Who decides what counts as a “frontier” or “high‑risk” model? Will NIST, CISA and the Pentagon coordinate so we don’t get eight different standards and a compliance circus? The smart play is narrow, technical review for models that demonstrably enable cyber or biological risks, done quickly and transparently. Call it defense-in-depth, not a permit office for AI. I’ll take a Trump White House that worries about national security over a Democratic panic that would kneecap American tech — but I won’t cheer a closed-door process that bakes in market winners. If Washington wants to keep Americans safe, it should be tough, clear, and fair — and it should never confuse safety with protecting incumbents.
