On May 12, 2026, the Colorado General Assembly passed SB 26-189, a bill that rewrites the state's landmark artificial intelligence law from scratch, covering mortgage underwriting, homeowner's insurance pricing, home inspections, and possibly AI-powered permit review. It takes effect January 1, 2027. And based on conversations with builders, inspectors, and real estate attorneys in Colorado and nationally, almost nobody in the residential construction industry has read it.
They should.
SB 26-189 creates the first state-level framework in the United States that explicitly requires companies to disclose when artificial intelligence materially influences a housing decision and to offer human review when the outcome is adverse, not a voluntary best practice or an industry guideline, but a law with enforcement authority vested exclusively in the Colorado Attorney General, who must provide 60 days’ written notice and an opportunity to cure before pursuing discretionary penalties through the Colorado Unfair Practices Act.
What the Law Actually Says
The 2026 Law replaces Colorado's original 2024 AI Act (SB 24-205), a sprawling piece of legislation modeled partly on the EU AI Act that regulated "high-risk AI systems" through mandatory impact assessments, bias audits, and risk-management programs. It never took effect: industry pushback, compliance cost estimates running into the millions annually, and a federal court stay stalled it before its June 30, 2026, effective date arrived.
What replaced it is narrower in mechanism but broader in its practical reach, regulating "automated decision-making technology" (ADMT) that uses personal data to make or assist "consequential decisions" concerning individuals, with definitions that align with California's privacy framework. The covered sectors: employment, education, housing, financial and lending services, insurance, healthcare, and public benefits, which means housing, lending, and insurance together form the residential construction industry's entire downstream ecosystem.
For companies that build AI tools — developers, in the law's terminology — the obligation is documentation: they must provide deployers with descriptions of intended uses, known limitations, and material updates. For companies that use those tools (deployers), the obligations cut deeper, requiring clear notice when ADMT materially influences a consequential decision, an explanation of the AI's role after an adverse outcome, and "meaningful human review and reconsideration" of the decision if a consumer requests it, provided it is "commercially reasonable."
That qualifier is doing a lot of work. The Attorney General's rulemaking, due by January 1, 2027, will define what "commercially reasonable" means in practice. Until then, every deployer in Colorado's housing market is operating on an educated guess about what level of human oversight the state will ultimately demand.
Where This Hits the Transaction
Consider a straightforward Colorado home purchase in early 2027. A buyer finds a listing with AI-enhanced photos. Gets a pre-approval from a lender using AI-assisted underwriting. Orders a home inspection from an inspector using Spectora's AI Report Assist, which matches voice observations to pre-approved template comments and saves 25 percent of inspection time across 10,000-plus users. Gets an appraisal from a lender relying in part on an automated valuation model, already regulated separately by the CFPB's AVM rule. Receives a homeowner's insurance quote priced partly by satellite AI that scored the roof condition. Gets a title search processed through an AI-driven title examination platform. Signs closing documents reviewed by an AI compliance engine.
Each of those steps involves a different company deploying a different AI system making or assisting a different "consequential decision" about the same buyer, and under SB 26-189, each company has independent disclosure obligations with no coordination mechanism, no standard format, and no shared infrastructure for delivering seven separate "AI was involved" notices to a buyer who is simultaneously trying to understand whether the foundation crack in photo 47 of the inspection report is a $200 cosmetic issue or a $40,000 structural problem.
That is not transparency — that is noise.
The Insurance Dimension Is the Urgent One
If you are a builder, inspector, or contractor reading this, you may be inclined to shrug: Colorado is one state, January 2027 is six months away, and the counterargument writes itself: this is a disclosure law, not a prohibition, the cure period is generous, and well-run companies documenting their AI use will barely notice the compliance cost.
All true, except the insurance market is not waiting for the AG's rulemaking. On January 1, 2026, Verisk released standardized AI exclusion forms CG 40 47 and CG 40 48 for commercial general liability policies. ISO forms underpin roughly 82 percent of U.S. property and casualty insurance policies, according to Cohen Seglias. Berkley, AIG, Great American, Philadelphia Insurance, Hamilton Select, Tokio Marine, Liberty Mutual, Chubb, The Hartford, Travelers, Berkshire Hathaway, Zurich: more than a dozen of the largest U.S. liability carriers have filed with state regulators to exclude AI-related liabilities from their policies. Regulators in most states are approving those filings.
Read that again: the tools are shipping, the law now requires disclosure of their use, and the carriers are stripping coverage for the liabilities they create, leaving the coverage gap squarely on the deployer.
Stanford Law School research found that even specialized legal AI tools hallucinate 17 to 33 percent of the time, with general-purpose models performing substantially worse. Spectora's own FAQ, to its credit, is unusually candid on the question: "Most E&O policies don't currently address AI use specifically, which doesn't mean you're protected by default. If you're using AI to generate findings, ask your carrier in writing how that's treated. A two-line clarification today saves a denied claim later."
That advice is worth more than most legal counsel charges for it.
Why This Matters Beyond Colorado
Colorado's 2026 Law matters nationally for the same reason California's building codes matter nationally: other states copy the homework.
At least six states had active AI regulation bills in the 2025-2026 legislative sessions, including New York (S.6278), Virginia (HB 697), California (SB 468), Illinois (HB 3021), Minnesota (SF 1886), and North Carolina (HB 970), and Colorado's rewrite gives them a working template. The original 2024 Act was a cautionary tale about overreach; the 2026 version shows that a disclosure-focused, enforcement-light approach can survive a hostile legislature, a skeptical governor, and a federal court challenge, so expect copycats by 2028.
Meanwhile, the federal vacuum persists: the CFPB's AVM rule covers algorithmic appraisals specifically, but beyond that, there is no federal framework governing AI's role in home inspections, permit review, construction lending, or insurance underwriting, and the states are filling the gap one by one while Colorado draws the map.
For builders and vendors selling AI-enabled tools to Colorado customers, the compliance timeline is tight. The AG's rulemaking starts immediately. Documentation requirements for developers are expected to be specific. If you cannot explain to a customer what your AI does, what data it uses, what its limitations are, and what happens when it is wrong, Colorado says that is not your customer's problem. It is yours.
What a Builder Should Do Before January
Inventory every AI tool in your workflow, not just the ones marketed as "AI," which may be none, but the ones that automate a decision affecting a person: the scheduling optimizer that prioritizes jobs, the estimating software that generates a bid, the inspection tool that matches observations to template comments, the underwriting engine your lender runs. Count them all.
Ask each vendor, in writing, three questions: what decisions does the AI make or assist, what personal data does it use, and what documentation will it provide under Colorado SB 26-189? If the vendor cannot answer, you have five months to find one that can.
Then call your E&O carrier and ask whether your policy covers liabilities arising from AI-assisted work product, and get the answer in writing. If the answer is no, or if the carrier cannot say, that is information you need before January, not after a claim.
If you think this does not apply to you because you are not in Colorado, consider that 75 percent of homebuyers already assume AI is embedded somewhere in the transaction, according to Cotality's 2026 AI in Housing report; they expect it, and they also expect someone to explain what it did. Colorado just made that expectation law, and the question is not whether other states follow but when.
What We Don't Know
Governor Polis signed SB 26-189 on May 14, 2026, so the law is real, but key ambiguities remain. The Attorney General's rulemaking will define critical terms, including "materially influence" and "commercially reasonable," and until those rules publish, the scope of disclosure obligations remains uncertain. The litigation challenging the original 2024 Act (x.AI LLC v. Weiser) is stayed, not resolved, and many of its constitutional theories may survive under the revised framework. The Attorney General has stated he does not intend to enforce the law until rulemaking concludes, which means the January 1, 2027, effective date may not be the real compliance deadline, but it remains the statutory one.
We have no data on compliance costs specific to Colorado's residential construction sector, and our "disclosure stack" analysis identifying seven potential AI notice points in a single transaction is illustrative, since some steps, particularly title search and closing review, may not qualify as "consequential decisions" under the final rules. We also could not confirm whether any state home inspection licensing board has issued formal guidance on AI use; Spectora's characterization that "some licensing boards are starting to question" AI-generated content is the strongest public signal we found, but it is secondhand.
The law is signed, the tools are shipping, the insurance market is pulling coverage, and the rulemaking starts this year. If you are waiting for certainty before acting, you are waiting too long.
Sources: Colorado SB 26-189 (2026); SB 24-205 (2024); SB 25B-004 (2025); Lexology analysis, May 2026; Mondaq, Baker Botts LLP; Verisk/ISO CG 40 47, CG 40 48 (eff. Jan 1, 2026); Cohen Seglias (ISO market share); Stanford Law School (AI hallucination rates); Spectora AI ethics FAQ and June 2026 product announcement; Cotality AI in Housing 2026 Report; ServiceTitan 2026 State of the Trades; NAHB/Wells Fargo HMI July 2025; CFPB AVM rule; Colorado Attorney General enforcement framework.