The RAISE Act Is a Start. The Rest Is Up to Us: Responsible AI Won’t Be Handed Down.
A Copernican Moment: Responsible AI, New York’s RAISE Act, and Why We Learn From the Ground Up
For fourteen centuries, the smartest people in the Western world believed the heavens revolved around the Earth — and when the observations didn’t fit, they didn’t question the center. They added epicycles: orbits within orbits, corrections upon corrections, an ever more elaborate machinery invented to preserve a wrong assumption. Copernicus didn’t discover new stars. He moved the center. And once the center moved, everything that had been impossibly complicated became simple.
We are living inside a Ptolemaic model of artificial intelligence. The assumption at its center: intelligence must radiate outward from a handful of hyperscale corporate hubs, and the rest of us — cities, cooperatives, school districts, small businesses, neighbors — exist to orbit, consume, and comply. When that model produces problems, we get epicycles: bigger data centers, longer terms of service, thicker compliance binders, another ethics board at another company we’ll never sit on.
This post is about the alternative — and about a piece of New York legislation that, read correctly, tells us exactly where the law’s reach ends and where our work begins.
What “Responsible AI” actually means
Strip away the branding, and Responsible AI is four old questions asked with new urgency:
Purpose. What human good is this system serving — and can whoever deployed it say so in one sentence a citizen would understand? A tool that can’t answer the purpose question isn’t neutral; it’s unaccountable.
Accountability. When an automated system harms someone, who answers? “The algorithm did it” is not an answer — it’s a plea of ignorance, and ignorance that could have been avoided (through testing, monitoring, documentation) has never excused anyone.
Judgment. Rules run out. Every algorithm is a general rule applied at machine speed, and every general rule eventually meets a case its authors never imagined. Responsible systems keep a human with real authority — not just presence — at the decision points that matter.
Redress. People wronged by automated decisions get notice, reasons, an appeal to a human who can decide differently, and repair. This is due process, and it is older than the phrase.
These questions are not new. They are, almost word for word, the architecture of Aristotle’s Nicomachean Ethics — which is why we’ve published a full knowledge base translating that 2,300-year-old framework into working AI policy tools at communityagents.ai/docs. What’s new is who gets to ask the questions, and of whom.
New York just moved: the RAISE Act
In 2025 and 2026, New York answered part of that question with law. The Responsible AI Safety and Education Act — the RAISE Act — passed the state legislature in June 2025, was signed by Governor Hochul that December, and reached its final form through a chapter amendment signed on March 27, 2026. It takes effect January 1, 2027, making New York the second state, after California, to enact comprehensive frontier AI safety legislation.
What it does, in plain terms: the law targets the developers of the very largest AI systems — companies with over $500 million in annual revenue building “frontier models” trained at enormous computational scale, with compute costs exceeding $100 million. Those developers must establish and publish safety protocols, and disclose serious safety incidents within 72 hours — a notably tighter clock than California’s 15-day window. A new oversight office inside the state’s Department of Financial Services gets rulemaking and enforcement authority.
Two things about the RAISE Act deserve real credit. First, it establishes the principle that the most powerful systems in history do not get to operate as black boxes — transparency and incident reporting are now legal obligations, not press-release promises. Second, it happened at the state level, against federal headwinds actively hostile to state AI regulation. That is federalism doing what it’s supposed to do: states acting as laboratories when Washington won’t.
But notice what the RAISE Act regulates: catastrophic risk at the very top of the stack. It governs a handful of frontier developers. It says almost nothing about the thousands of ordinary decisions — in county benefits offices, school districts, small businesses, city procurement departments — where AI will actually touch most people’s lives. That’s not a flaw in the law. It’s the boundary of what law aimed at the center can do.
The Copernican turn
Here is the reorientation. The question is not only “how do we constrain the center?” It is “what if the center is in the wrong place?”
The Copernican revolution didn’t just redraw the sky — it redistributed the authority to describe reality. Within a generation, observation and reason mattered more than proximity to established power. The parallel for our moment: the capability to understand, govern, deploy, and own AI does not have to live exclusively in a few corporate hubs, with everyone else waiting for terms of service. It can live in communities — the same way municipal broadband proved that fiber doesn’t have to come from a telecom monopoly, and the way community solar proved that power doesn’t have to come from a distant plant.
But capability, unlike regulation, cannot be legislated into existence. It has to be practiced into existence. Aristotle’s oldest insight applies: we become capable of a thing by doing it, together, repeatedly. No agency memo makes a community fluent in AI governance. A monthly deliberation practice does. No vendor certification teaches a caseworker when to override an algorithm. Experience, shared and compared with peers, does.
Which means the essential Responsible AI infrastructure of the next decade isn’t only legal or technical. It’s pedagogical: people teaching each other, laterally, the concrete techniques of systemic change. Running a structured deliberation before a deployment decision. Asking the procurement questions that make vendor black boxes contractually disqualifying. Standing up community-owned agents on community-owned infrastructure — increasingly, renewable-powered infrastructure we control — instead of renting intelligence by the token. Governing local data as a commons, with the rules Elinor Ostrom showed communities have always been able to write for themselves.
That is what “AI that works for us, and not over us” means in practice. Not a slogan — an operating model. Systems whose purpose we set, whose failures we can see, whose decisions we can appeal, running on infrastructure we own, maintained by skills we hold in common.
The center is where we stand
The RAISE Act watches the frontier. Good. Someone must. But the frontier was never going to be where this is decided. It will be decided in the middle of the stack and at the ground level — in whether communities build the shared fluency to shape these tools before these tools shape them.
Copernicus worked mostly alone, and it took a century for the world to catch up. We don’t have a century, and we don’t have to work alone.
That’s exactly why we’ve launched a hands-on certification cohort — practitioners teaching practitioners the working techniques of community-centered AI: agent-building, deliberation practice, procurement strategy, data governance, and the infrastructure to run it all ourselves.
Also learn Open Source models in Project Based Learning for your team.
Bring your community’s hardest problem. Leave with a technique someone else already tested — and teach one of your own. The center moves when we move it.
Further reading: our full knowledge base, “Aristotle & Responsible AI,” is free at communityagents.ai/docs — including model policy language, a deliberation workshop guide, and talking points for legislators.
