I watched a slim, typed letter land on a desk in Washington and felt the room go colder. You can sense the calculation: pastors, strategists and Steve Bannon signing a single demand. I will take you through what they want, who they are, and why this matters more than most headlines admit.
At a crowded press briefing, allies pressed their names to a single page — who signed the letter and why it matters
I’ve tracked political petitions for years, and this one reads like a dossier and a prayer rolled into one. Steve Bannon and more than 60 Trump allies — a mix of campaign veterans, conservative operatives, and an unusual number of pastors — asked President Trump to issue an executive order that would force federal review of frontier AI models before public release.
The group behind the push, Humans First, frames the ask as national-security hygiene: mandatory government vetting of anything deemed a potentially dangerous AI. They name risks to elections, finance, critical infrastructure, biosecurity and military systems — a list meant to make anyone who cares about the machinery of the state sit up.
At the White House lawn, the administration’s instincts were visible — how the Trump team has treated AI so far
You’ve seen the pattern: the administration favors light regulation when the private sector promises growth and job headlines. Earlier this year, Trump signed an executive order aimed at restricting state-level limits and protecting industry interests. That order was framed as protecting national and economic security while blocking what the White House called burdensome regulation.
So this new push from Bannon and his allies runs against that grain — a conservative coalition asking for stronger federal gatekeeping of technology companies that historically enjoy GOP-friendly treatment.
At a Pentagon negotiating table, tensions over Anthropic became a real-world test case — what happened with Anthropic and Mythos
I sat through briefings where Pentagon negotiators and AI firms parsed language about surveillance and weaponization. The standoff with Anthropic began when the Department of Defense worried the company would not allow military use consistent with U.S. needs; Anthropic pushed back and then filed lawsuits after being labeled a supply-chain risk.
The company’s latest model, Mythos, has been distributed in a tightly controlled program to select companies, governments and organizations to strengthen cybersecurity. That limited release flipped a switch in the debate: the administration considered dropping its public rebuke of Anthropic while negotiating agreements with other firms like Google, Microsoft, and OpenAI to let the government review models before they reach broad audiences.
What is frontier AI and why is it risky?
Frontier AI refers to models whose capabilities approach or exceed thresholds where the unknowns are large and the potential for systemic harm rises. You can think of these systems as tools that scale problems as fast as they scale solutions — capable of reshaping information flows, automating complex decisions, and interacting with critical infrastructure in ways ordinary software cannot.
At a church gathering and a strategy room I observed the same tone — moral anxiety blends with political calculation
One striking detail: many signers are pastors. That mixes ethical alarm with political muscle. When clergy and party operatives converge, the framing shifts from technical regulation to a question of trust and control over technologies that affect souls as well as systems.
The letter also singles out company executives and “unelected elites” experimenting on the public without democratic checks — a phrase chosen to land with populist audiences as much as with Beltway policymakers.
Can the President require AI companies to get federal approval?
Yes and no. The President can direct federal agencies and use procurement and classification authorities to limit government use and influence private behavior. But a broad, pre-release vetting regime would require legal scaffolding, interagency coordination, and buy-in from firms like Anthropic, Google, Microsoft and OpenAI — or else face litigation and political backlash.
At a policy workshop, the mechanics were hashed out — what the proposed executive order would do
The letter asks for mandatory testing and government approval of advanced models before public deployment. Practically, that would mean new review protocols, disclosure from companies about capabilities and training data, and likely a playbook for when models are deemed too risky for public release.
Those demands aim to flip authority from private firms to federal hands: no single company should be able to release technology that could threaten national security without a government stamp of approval, the letter argues.
There are two ways to look at this. To some, it’s a necessary public-safety measure; to others, it’s political theater that hands the state a chokehold on innovation. The choice will shape who controls the levers of technological power — boards of directors or elected offices and their agencies.
Regulatory pushbacks and industry pacts are already in motion: the executive branch has entertained agreements with firms to inspect models, and the Anthropic episode shows how negotiations can freeze into litigation and then thaw again. The tug-of-war is live, and the players include not just firms and officials, but foreign governments, venture capital, and academic labs.
The debate matters because the consequences are concrete — attacks on infrastructure, manipulated elections, or bio risks could shift markets, life choices, and national safety in ways that are hard to reverse. The question the letter forces onto Trump’s desk is whether you trust private companies to police themselves or want federal gates that slow releases down and put reviewers in charge.
One more observation: these models are now part public commons and part proprietary fortress, a duality that makes policy messy. They are like a loaded clock whose hands keep moving toward a deadline.
Another truth I’ve learned watching tech and politics collide is that power often follows the path of least resistance — corporations will press for access, governments will press for control, and the public will be asked to accept outcomes, good or bad. The system feels like a dam with hairline cracks, and every model release tests whether policymakers patch or pry the fissures wider.
If Trump signs a restrictive executive order, it will rewrite the rules for Anthropic, Google, Microsoft, OpenAI, and the rest — and set a precedent for future administrations. If he declines, the conservative coalition that put this on his desk will either escalate or reframe its demands. Which do you think will win: the firms with global reach or the political coalition trying to pull the plug?