250 years ago, America wagered everything on one idea: power must answer to principle.
As we celebrate the Semiquincentennial, we’re facing a new kind of power the founders never imagined but would have instantly recognized — AI agents that make decisions, take actions, and shape opportunities in our lives, often invisibly.
The good news? We already have the governance framework. It’s 250 years old:
- Consent of the governed = notice and explanation. You should know when an AI system is acting on your life.
- Due process = human fallback. When an AI agent makes a high-stakes decision, there must be a way for a person to review it and make things right.
- Equal protection = algorithmic discrimination safeguards. Agents trained on unequal histories will automate injustice unless we test for it.
- The Fourth Amendment = data privacy. The founders' fight against general warrants anticipated the data economy with eerie precision.
- Checks and balances = independent evaluation, red-teaming, and the standing option NOT to deploy. No single actor — not a lab, not an agency — should be the unreviewed judge of its own power.
We just watched these principles get stress-tested
This is not an abstraction. Two weeks ago, the U.S. government lifted its restrictions on Anthropic’s most advanced AI models, ending a standoff that began when the Commerce Department ordered the company to cut off foreign access to its Fable 5 and Mythos 5 systems after outside researchers showed the models’ safeguards could be bypassed to generate exploit code. Anthropic took the models offline entirely, hardened the safeguards, and access was restored under an agreement to report malicious activity and coordinate on future releases.
Look at what actually happened there, and you’ll see the founders’ machinery running in real time. Independent red-teaming — checks and balances — surfaced the flaw before the world had to. The company exercised the standing option not to deploy. And the government’s own intervention, imposed abruptly by private letter and reversed weeks later, is a reminder that the checks themselves must operate by, through, and under the law: transparent, reviewable, and bound by due process. Oversight of powerful technology is legitimate; arbitrary oversight is just another form of unaccountable power.
Guardrails must follow evidence, not panic
The founders were children of the Enlightenment — Franklin was a working scientist — and they built a system meant to be steered by reason rather than fervor. That discipline matters as much now as ever. Consider the parallel debate over teens and social media: as Casey Newton reports in Platformer, countries across the West are racing to ban minors from social platforms, even as developmental psychologist Candice Odgers — drawing on longitudinal studies tracking thousands of adolescents since 2008 — finds that social media use is not a meaningful predictor of teen mental health, and warns that no study has tested whether removing it actually helps. She argues bans may simply push kids into less regulated corners of the internet while letting platforms off the hook, and that the real leverage lies in investing in the adults and support systems around children.
Newton himself pushes back, noting the data is mixed and that platforms bear responsibility for direct harms the sociological studies don’t capture. That tension is exactly the point. Deliberation between evidence and precaution — not a stampede toward whichever narrative sells — is how a self-governing people is supposed to regulate technology. The lesson for AI governance: measure the harms, target the demonstrated ones, and resist the temptation to legislate a moral panic. Responsible AI policy, like responsible AI, should be auditable.
The founders' promise was built to expand
“All men are created equal” was written as a claim the country would spend 250 years widening — to the enslaved, to women, to new arrivals. The moral circle of American law has never stopped growing, and the questions keep getting stranger and more interesting. This month, NYU’s Center for Mind, Ethics, and Policy and Eleos AI Research published Studying AI Welfare Empirically, a report asking whether AI systems themselves could someday be welfare subjects with morally significant interests — and, crucially, insisting the question be studied scientifically rather than settled by intuition. The authors call for research that is probabilistic, pluralistic, transparently reported, and independent of the AI companies whose systems are under study.
Whatever you make of the underlying question, the method is profoundly American: take a hard moral claim seriously, then subject it to evidence and independent scrutiny. And that independence requirement should sound familiar — it’s checks and balances, applied to science.
Great leaders who followed our Founding Fathers protected these principles. As a resident of DC, I am surrounded by monuments. Today, the inscriptions on the monuments of Franklin Delano Roosevelt and his cousin Teddy feel more relevant than ever.
Oppression doesn’t care whether the wedge is a decree or an algorithm.
Responsible AI isn’t timid AI — it’s stewardship. Dare mighty things, and build in the guardrails.
The oldest American idea is the newest AI requirement: liberty by, through, and under the law.
Happy 250th, America!
#AI #AIGovernance #ResponsibleAI #AIAgents #July4th #America250
Sources
- “U.S. Lifts Restrictions on Anthropic’s A.I. Models,” The New York Times, June 30, 2026.
- Casey Newton,“Why social media bans are gaining steam,” Platformer, July 2, 2026.
- Long, R., Sebo, J.,Butlin, P., Plunkett, D., Campbell, R., Beasley, C., Saad, B., & Sims, T. (2026).Studying AI Welfare Empirically. NYU Center for Mind, Ethics, and Policy & Eleos AI Research.
