This month marked a rare moment of self-restraint in Silicon Valley. After months of escalating warnings, the four largest AI labs publicly endorsed slowing their model development. It began Saturday, September 12, when Anthropic’s Dairo Amodei urged a more deliberate pace; Sam Altman and Elon Musk agreed within hours, and the four largest labs found themselves briefly united in admitting that their own products may be moving faster than anyone’s ability to govern them.
Washington’s response was to argue about jurisdiction, not to deny the problem. Speaker Mike Johnson conceded that Congress is less qualified in pushing this “frontier” and must “resist jumping in and imposing some sort of emergency moratorium. It’s got to be done right, safely, wisely.”
President Trump struck a similar note, arguing that ceding America’s AI lead poses a greater danger than moving too fast. This is a tradeoff that his administration will have to continue weighing as this debate continues.
The real disagreement is inside the conservative coalition, and it’s about which level of government gets to hold the pen. Trump administration officials and Sen. Ted Cruz warned that allowing regulations would slow innovation and ultimately cede the race to China. Fellow Republicans, like Sen. Josh Hawley, Sen. Marsha Blackburn, and Gov. Ron DeSantis, helped strip a federal moratorium on state AI laws from the 2025 budget bill in a 99-1 rout. Rep. Majorie Taylor Green put it bluntly on X: “Federalism must be preserved.”
Both sides are defending a real conservative principle. Both stop the argument one rung too high.
The principle is subsidiarity. As Pope Pius XI observed in Quadragesimo Anno:
[I]t is an injustice… to assign to a greater and higher association what lesser and subordinate organizations can do.
The federalists are right that a national rulebook stripping states of their police power is centralization wearing a deregulatory costume. The nationalists are right that the notion AI techs can “police” themselves is not subsidiarity but abdication. But neither camp finishes the descent it claims to believe in. Below the congress is the statehouse, below the statehouse is the local community, and below all of this is the household, which is where children are actually encountering these tools.
In July 2024, the Senate passed the Kids Online Safety Act (KOSA), though it died in the House. In 2026, lawmakers advanced the GUARD Bill, restricting minors’ interactions with AI “companions” after chatbox manipulation was implicated in suicide, violence, and sexualization.
Blackburn’s 291-page “TRUMP AMERICA AI Act,” released as a discussion draft in March 2026, would carry much of this into federal law, while preempting conflicting state rules. The administration’s National Policy Framework, out days later, urged age assurance and parental controls over privacy settings and content exposure.
Here’s the potential problem with that: Every one of these measures assumes a parent on the other end who knows what to do. That parent largely does not exist yet.
The White House’s 2025 executive order “Advancing Artificial Intelligence Education for American Youth” addresses students and teachers. The 2026 updated “Protecting Children and Empowering Parents” section offers parents dashboards. But a dashboard is not formation. Parents cannot use safeguards they do not understand. Many either fear AI irrationally or ignore it entirely. Both dispositions are not productive for educating their child.
The consequences are already apparent. Children receive school-issued laptops as early as 4th and encounter pornography by age ten. AI companions, algorithmic tutors, and generative content now shape their imaginations before parents even know these tools exist.
So here is the rung nobody is standing on. Parental AI literacy will not come from a federal curriculum, and we should be glad of that.
It belongs to the institutions that already have parents in the room every week: school boards, homeschool co-ops, diocesan and denominational family-life offices. These bodies are part of the formation of the children in their communities. They are the natural venue for a two-hour evening that teaches a mother what an AI companion is, why her son’s chatbox remembers his loneliness, how algorithmic recommendations could never replace parental advice, and what questions to ask her son after school.
None of this requires a new federal office, a compliance regime, or another 291-page bill. States can fund exactly this in the same way that they already fund libraries and vocational partnerships. A state legislature that appropriated modest grants for parent-facing AI literacy, administered through existing local institutions rather than a new agency, would be doing exactly what subsidiarity requires: furnishing help to the lesser association rather than absorbing it.
This is not a novel idea so much as a very old one. Aristotle argues that moral formation begins in the household, Aquinas teaches that the family is more fundamental than the state, and Augustine, recalling his own childhood, insists that parents bear the first responsibility for shaping the loves of their children. None of them expected the city to do this. All of them expected the city to make it possible.
As C.S. Lewis wrote in The Abolition of Man, “We make men without chests and expect of them virtue and enterprise.” A decade of technology policy has been an argument about restraint: who imposes it? At what level? On whom?
Restraint is necessary and insufficient. A guardrail keeps a car on the road; it does not teach it to drive.
Pandora’s box has been opened. Congress and the White House can keep arguing about who writes the rules. The rest of America, in the meantime, has their own rung to occupy. That’s the one closest to the family, which no one in Washington is contesting and no one outside it has yet claimed.
