Trump's AI Force vs. labs quietly slowing down

On Saturday, September 19, 2026, Trump announced on Truth Social that he would form an "AI Force," writing: "For this purpose, I am forming the AI Force, much like I did Space Force, which has been a tremendous SUCCESS." He also said he would appoint a new AI czar, adding that "only High I.Q. individuals need apply" for the job, according to NBC News.

The post gave no budget, no staffing plan, no legal authority and no timeline for naming that czar, Al Jazeera reported. What it did make explicit is the administration's stance on regulation: none. "We will not in any way hinder or stifle the Growth of this incredible Industry. Rather, we will cherish it, help it, and watch over it, as it grows!" Trump wrote, per NBC News. He dismissed existential-risk warnings about AI as a "hoax," saying bad conduct would be handled through existing criminal and civil law rather than any AI-specific rulebook, and framed the whole effort as a competitive play: "We are leading China, and the rest of the World, and I intend to keep it that way!"

That framing is why the story spread as fast as it did. Within a day it had been independently covered by NBC News, CNN, The Washington Post, Al Jazeera, Axios, Fox News, Fortune and Salon — an unusually wide same-day spread for a single social media post about AI policy. Part of what made it land was timing: it arrived hours after two of the labs building the systems in question signaled they wanted to go slower, not faster.

The labs are moving the opposite direction

While the White House was promising to "cherish" AI growth with zero new constraints, Anthropic CEO Dario Amodei was introducing new internal safety measures and a slower development pace, and OpenAI CEO Sam Altman was reportedly floating the idea of a cross-lab pact among leading labs to deliberately decelerate development, Fortune reported. Google DeepMind co-founder Demis Hassabis has also voiced support for slowing down, according to the same report.

Inside Anthropic specifically, the tension is not abstract. Researcher Evan Hubinger said publicly, "we really do earnestly believe AI could kill all humans," per Fortune's reporting. Around the same time, another Anthropic researcher, Jacob Coxon, resigned and accused both Anthropic and OpenAI of building increasingly powerful systems irresponsibly. That is a safety researcher quitting a safety-focused lab over its own pace of development — in the same week a sitting president called safety concerns a hoax.

It's worth noting this isn't the administration's first AI-czar-shaped role. David Sacks previously held an AI-czar-style advisory position for Trump before resigning earlier in 2026; he now chairs Trump's Council of Advisors on Science and Technology instead, per Al Jazeera. An "AI Force" with an unnamed czar is, at minimum, the second attempt at giving this administration's AI posture an institutional face.

Public sentiment doesn't obviously back the hands-off approach either. The U.S. House of Representatives passed a bill 417-3 opposing the community impact of AI data centers, a near-unanimous vote that reflects bipartisan unease running counter to the White House's growth-first message, Al Jazeera reported. Whatever the AI Force turns out to be, it is not responding to a public mandate for less oversight.

Why this matters if you build with these models

Strip away the theater and the practical signal for engineering teams is simple: federal policy is not going to define your AI safety guardrails any time soon. No agency is being described, no rulemaking process, no enforcement mechanism beyond "existing criminal and civil law." That leaves exactly the actors who were already setting the norms — the labs themselves — as the only source of anything resembling a safety floor.

That's a meaningfully different picture depending on which lab you build on. Anthropic is publicly tightening internal safety measures and slowing its own pace. OpenAI's CEO is reportedly discussing a voluntary, multi-lab deceleration pact rather than waiting for regulation to force one. DeepMind's co-founder has said something similar in public. None of this is legally binding, and none of it is verified by any outside body — it's lab leadership talking about lab leadership's own choices. But it's also the only signal on the table right now, since the government has explicitly declined to set any technical or safety standard.

For a team choosing which model provider or which lab's safety commitments to lean on, that means doing the diligence yourself instead of assuming a regulatory backstop exists. Read what each lab actually publishes about its internal safety process, not just its marketing. Track researcher departures and public statements from safety teams — Coxon's resignation and Hubinger's comments are exactly the kind of signal that tells you more about a lab's internal risk tolerance than any press release does. If a lab is willing to publicly commit to a slower release cadence or a cross-lab pact, treat that as a data point worth weighing against raw capability benchmarks, especially for anything you're putting in front of production traffic without a human in the loop.

What to actually do about it

Don't wait for an "AI Force" briefing to tell you what's safe to deploy — there's no indication one is coming, and the announcement itself contained no operational detail to act on. Instead, build your own vendor-risk process around lab-published safety commitments: read Anthropic's and OpenAI's safety documentation directly rather than secondhand summaries, watch for concrete commitments (deceleration pacts, external red-teaming, disclosed incident reports) versus vague reassurance, and weight departures from safety teams as a real signal. If your organization has any AI governance policy at all, this is the moment to make sure it doesn't quietly assume federal rules will backstop it — right now, they explicitly won't.

Keep an eye on two threads going forward: whether Trump actually names an AI czar and what authority that role gets, and whether the OpenAI-reported cross-lab slowdown pact turns into anything concrete. Either one would change the calculus described here — the first by giving the government an actual lever, the second by giving the labs a shared, harder-to-walk-back commitment. Until then, the safety net you're relying on is whichever lab you chose to build on.