The AI Rulebook Washington Can't Pass
Randy Hall, CEO

Federal lawmakers have tried twice in two years to lock in one national AI standard and failed both times, most recently when a bipartisan discussion draft stalled before reaching introduction. States legislated anyway. Colorado's rewritten AI law sets compliance deadlines for January 1, 2027, and more states will follow. Waiting for Washington to settle this is not a strategy.
Why hasn't Washington settled this already?
The first real attempt collapsed in public. In July 2025, the Senate voted 99-1 to strip a decade-long moratorium on state AI laws out of that year's budget reconciliation bill, after a last-minute deal on a softer version fell apart. That vote told you almost everything about the odds of a clean federal override arriving soon. Nearly every senator who voted wanted the states to keep their own authority over AI, regardless of party.
The second attempt is softer and still unfinished. Representatives Jay Obernolte and Lori Trahan released a discussion draft of the Great American AI Act on June 4, 2026, built to collect feedback before anyone formally introduces it as a bill. Instead of repeating the sweeping ban senators already killed, the draft's Section 121 would preempt state and local laws that specifically regulate AI model development for three years, leaving deployment, distribution, and use rules to the states. A narrower, time-limited carve-out is an easier sell than what the Senate already rejected, but it is still a draft with no sponsor, no committee vote, and no floor schedule. Build your plans on what exists, not on what a discussion draft might become.
Both failures share a pattern worth noticing. Opposition to a federal override has not split cleanly along party lines in either case. Lawmakers from states that already passed their own AI rules, red and blue alike, keep resisting a Washington standard that would strip out what their own legislatures already built. That dynamic is exactly why betting your compliance roadmap on a single national rulebook arriving on any predictable timeline is a weak plan for a CEO to carry into next year's budget.
What does the state patchwork look like right now?
It is getting bigger, not smaller. States had enacted 109 AI laws across 29 states by July 1, 2026, a pace slightly behind the 121 laws enacted by the same point in 2025, according to a running tally kept by Tech Policy Press. A slower pace is not the same as a slowing trend. States that already have a law on the books are rewriting and expanding it, which is exactly what happened in Colorado this year.
That matters for a reseller because most of these laws define who they reach in terms that have nothing to do with where a product was built. A company does not need an AI research lab to count as a "deployer" under most of these statutes. It needs to put a system that makes or informs a decision in front of a customer, under its own name. That describes a white-labeled AI tool running under your brand about as well as it describes anything else these laws were written for.
Consider a hypothetical MSP running the same AI-powered client portal across accounts in five different states. The underlying system never changes, but that MSP could still end up tracking five separate disclosure formats, five different adverse-decision notice windows, and five unrelated recordkeeping clocks, with no guarantee any of them line up. The real cost here is rarely the technology itself. It is the paperwork trail a reseller has to rebuild behind it, once per state, for as long as the patchwork keeps spreading.
What is Colorado's rewritten law actually asking of you?
Colorado scrapped its original 2024 AI Act before it ever took effect and replaced it with a narrower, disclosure-focused law. Governor Jared Polis signed SB-26-189 on May 14, 2026, the bill Colorado titled Automated Decision-Making Technology. The law itself takes effect January 1, 2027. The law covers "covered ADMT" used to materially influence consequential decisions across employment, housing, lending, insurance, health care, education, and essential government services, which is a wide net for any MSP selling AI tools into those client verticals.
The split in who owes what is the part worth studying closely. Developers have to hand deployers technical documentation covering intended use, training data categories, known limitations, and guidance on human review, plus notice when something material changes. Deployers, meanwhile, have to tell consumers clearly when they are dealing with the technology and explain in plain language, within 30 days after the decision, what role the technology played. Developers and deployers alike also have to keep records showing compliance for at least three years after the decision, and the law gives violators a right to cure that expires at the start of 2030, with the Colorado Attorney General enforcing the framework under the state's consumer protection law.
If you resell an AI tool under your own name, you are very likely sitting on the deployer side of that line even when your vendor wrote every line of the model's code. That obligation does not show up automatically in a reseller agreement unless someone goes looking for it.
The law also sets a baseline for dividing fault between developers and deployers when an automated decision ends up in a discrimination lawsuit, even though it creates no new private right of action on its own. That detail matters more than it sounds like it should. If a client ever sues over a biased automated decision your white-labeled tool helped make, the contract between you and your AI vendor, not the statute alone, decides who actually absorbs the cost. Reading that contract now is cheaper than reading it for the first time during a deposition.
So what should you actually build before the next deadline?
Stop treating each new state law as a one-off compliance project. With federal preemption stuck at a vote nobody can win and a discussion draft nobody has introduced, the realistic planning horizon is a patchwork that keeps growing for years, not a single rulebook that settles the question. The practical move is to build one documentation and disclosure practice that is strict enough to satisfy the toughest state you operate in, then apply it everywhere. In practice, that one practice has to cover the same ground regardless of which state is asking:
- Technical documentation of intended use, training data categories, and known limitations for every AI tool you resell
- A disclosure consumers actually see before they interact with the system, not one buried in a terms-of-service page
- A human review path for anyone who receives an adverse automated decision
- A records retention clock set to your strictest state, not your most lenient one
NIST's AI Risk Management Framework is intended for voluntary use, not built around any single state's statute. It organizes AI governance around four functions: Govern, Map, Measure, and Manage. That voluntary, state-agnostic design is exactly why it works as a common backbone no matter which law applies to you this year. Adopting that structure internally does not satisfy Colorado's statute by itself, but it gives you one documentation habit to maintain instead of fifty, and it reads as diligence to a regulator or a client's legal team no matter which law eventually applies.
None of this is free, and building it alone gets more expensive every time another state passes its own version of Colorado's law. Getting your team fluent in what developers and deployers actually owe each other under these statutes, instead of re-learning it after the next state passes its version, is squarely what Forge University exists to do for MSPs who would rather train once than scramble every legislative session. If you are weighing how much of this compliance groundwork to build in-house versus buy into your stack already handled, the stack builder walks through that trade-off tool by tool, and the broader product catalog shows what a vendor can already be carrying for you before the next deadline hits.
The regulatory horizon for the channel is not ambiguous anymore. Washington cannot agree on an override, states will keep legislating into 2027 and beyond, and the MSPs who treat documentation as a product feature rather than an afterthought will spend less time rebuilding compliance every time a legislature reconvenes. See the full stack to see where that discipline is already built in.
