Three Moving Pieces, One Calendar
US AI regulation in 2026 is not one thing — it is three overlapping fronts, each with its own risks. For compliance, product, and legal leaders running AI features in multiple US states, the picture looks like this:
- California SB 53 — “Transparency in Frontier Artificial Intelligence Act” — effective January 1, 2026.
- Colorado AI Act (SB 24-205) — a broad algorithmic-discrimination statute — effective June 30, 2026 (delayed from February 1, 2026 via SB 25B-004).
- Federal executive order of December 11, 2025 — “Ensuring a National Policy Framework for Artificial Intelligence” — launches a DOJ-led AI Litigation Task Force to challenge state laws on preemption grounds.
These move on different tracks, but they interact. A team shipping an AI product into the US market in 2026 needs a stance on each one — and should not assume the federal push will erase state obligations before the deadlines hit.
California SB 53: Transparency for Frontier Developers
California Governor Gavin Newsom signed SB 53 — the Transparency in Frontier Artificial Intelligence Act — on September 29, 2025, with an effective date of January 1, 2026. Per the WilmerHale analysis, the law targets a specific class of developers:
- “Large frontier developers” — those with annual gross revenue exceeding $500 million.
- Building “frontier models” — foundation models trained using more than 10²⁶ integer or floating-point operations.
These developers must publish a risk management framework, produce transparency reports, and disclose certain information about their model training and safeguards. SB 53 is narrow — it does not regulate every AI system — but it hits the companies most visible in the generative-AI market.
Several other California AI laws also took effect on January 1, 2026, including AB 2013 (training-data disclosure for generative AI developers) and SB 243 (chatbot disclosure rules). The King & Spalding roundup treats January 1, 2026 as “the new year, new rules” moment for AI in California.
Colorado AI Act: Delayed, Not Dead
Colorado’s SB 24-205 — the Colorado AI Act — was the first comprehensive US state AI law, signed in 2024. Its stated goal: require developers and deployers of high-risk AI systems to use reasonable care to protect consumers from algorithmic discrimination.
The original effective date was February 1, 2026. During a special legislative session in August 2025, Colorado passed SB 25B-004, pushing the effective date to June 30, 2026. Baker Botts’ implementation alert explains that the delay gives developers, deployers, and the Colorado Attorney General more time to finalise operational details — not a reprieve from the substance.
When it kicks in, the Colorado AI Act will require:
- A documented risk management programme for high-risk AI.
- Impact assessments covering known or reasonably foreseeable risks of algorithmic discrimination.
- Notice to consumers when they are subject to a consequential decision made by a high-risk AI system.
- A right to appeal or correct discriminatory outcomes.
- Reporting to the Attorney General where discrimination is discovered.
For enterprises deploying AI in employment, housing, lending, insurance, education, or essential services, this is the most operationally demanding US state AI regime to date.
Advertisement
The December 11, 2025 Executive Order
On December 11, 2025, President Trump signed an executive order titled “Ensuring a National Policy Framework for Artificial Intelligence”. As summarised by Sidley Austin and Morgan Lewis, the order does three things that reshape the compliance landscape:
- DOJ AI Litigation Task Force — the Attorney General is directed, within 30 days, to stand up a task force whose sole responsibility is to challenge state AI laws inconsistent with the order’s policy, on grounds of unconstitutional interstate-commerce regulation or federal preemption. California’s transparency laws and Colorado’s algorithmic-discrimination statute are named explicitly.
- Federal agency pressure — FCC to explore a federal reporting and disclosure standard, FTC to issue a policy statement on how deceptive-practices law applies to AI, and Commerce to evaluate state laws that “conflict” with federal AI policy by March 11, 2026.
- Federal funding conditions — review of potential conditions on federal funds flowing to states, intended to discourage “onerous” state AI regimes.
Crucially, an executive order is not itself preemption — it can launch litigation and pressure, but state laws remain in force until a court rules. The EPI policy tracker and NPR’s legal analysis both stress that the legal outcome is uncertain.
What Multi-State AI Compliance and Product Teams Should Do Now
The three-front reality does not offer a wait-and-see option. The June 30, 2026 Colorado deadline is 10 weeks away; California SB 53 is already in force; and the DOJ task force filing timeline is active. These four actions address each front without duplicating work.
1. Build the Documentation File Once and Reuse It Across All Three Tracks
The common denominator across California SB 53, the Colorado AI Act, and any future federal framework is documentation: model architecture, training data summary, evaluation results, intended use, and risk controls. Morgan Lewis’s April 2026 analysis confirms that documentation-first teams consistently pass multi-state reviews faster because the same dossier satisfies the transparency report (SB 53), the risk management programme (Colorado), and the downstream-provider information package required by the EU AI Act for GPAI models. Build the file in a format that can be updated incrementally — a versioned document store with named sections, not a PDF — so that each new state deadline triggers an update rather than a rebuild.
2. Treat Colorado June 30 as a Hard Deadline, Not a Soft Target
Colorado SB 25B-004 delayed the effective date to June 30, 2026, but as Baker Botts’ implementation alert stresses, the delay was procedural, not substantive. The high-risk AI definition — AI systems that make consequential decisions in employment, housing, lending, insurance, education, or essential services — covers a wide range of enterprise AI products. Teams that have not started the impact assessment template, consumer notice, and appeal-mechanism design should treat every remaining week as a critical-path item. The Colorado Attorney General’s enforcement posture after go-live is the biggest unknown; early compliance avoids being the first test case. Two to three weeks of sprint work is enough to clear the minimum viable compliance bar for most mid-market AI products.
3. Add SB 53 Pass-Through Clauses to Vendor Contracts Now
If your product is built on a frontier model from OpenAI, Anthropic, Google, or Microsoft, and your vendor meets the SB 53 threshold ($500M annual revenue, models above 10²⁶ FLOPs), you should expect pass-through disclosure obligations. WilmerHale’s SB 53 analysis explicitly flags that downstream deployers will receive contractual pressure to represent their intended-use documentation to the frontier provider. Updating vendor agreements during the next contract renewal to include an SB 53 disclosure schedule — stating intended use, known deployment contexts, and risk controls applied — protects both parties and shortens future compliance reviews.
4. Monitor DOJ Task Force Filings, Not Press Coverage
The December 11, 2025 executive order launched a task force whose enforcement actions will be visible in DOJ public filings and federal court dockets, not in press releases. The EPI policy tracker and Sidley Austin’s analysis both note that the legal outcome on preemption is genuinely uncertain — courts must rule, and the Commerce Department’s state-law assessment is scheduled for March 11, 2026. For practical compliance purposes, the task force changes nothing before a court ruling. Compliance teams should set a quarterly review to check active DOJ filings and FCC/FTC policy statements rather than reacting to every news cycle. Colorado and California laws remain in force until a court says otherwise.
The Regulatory Question
The December 11, 2025 executive order’s DOJ AI Litigation Task Force is attempting to answer a constitutional question that the United States has never formally resolved for technology regulation: can a state impose compliance obligations on AI systems that affect interstate commerce, or does the federal government’s authority over that commerce preempt the field? California and Colorado believe their laws are constitutional. The Trump administration believes they are not. The answer will come from federal courts, probably on a multi-year timeline, and could take different forms for different provisions — a ruling that strikes down Colorado’s algorithmic-discrimination notice requirement might leave California’s SB 53 transparency obligations untouched, or vice versa.
What makes this regulatory question structurally important for AI teams beyond the US is that its resolution will set a template for how other federal systems — the European Union’s member state implementation of the EU AI Act, Canada’s proposed AI and Data Act, and Australia’s evolving framework — handle the tension between national-level AI policy and jurisdiction-specific enforcement. A US court ruling that federal preemption of state AI law is constitutional would signal that centralized AI governance frameworks are the viable architecture; a ruling in favor of state authority would confirm the fragmented multi-jurisdiction compliance reality that already characterizes privacy law. For enterprise AI compliance leaders building documentation frameworks in 2026, the preemption outcome does not change the immediate obligations — Colorado’s June 30 deadline and California’s January 1 effective date are real regardless — but it will shape whether the 2027 documentation architecture is built for convergence or for persistent fragmentation.
Frequently Asked Questions
Is the Colorado AI Act still happening?
Yes. Colorado SB 25B-004 delayed the effective date of the Colorado AI Act from February 1 to June 30, 2026, but kept the substance in place. Companies with high-risk AI systems in Colorado should treat June 30, 2026 as the real deadline and finish implementation before it.
Who does California SB 53 actually apply to?
SB 53 targets “large frontier developers” — those with annual gross revenue above $500 million — building foundation models trained with more than 10^26 integer or floating-point operations. Most mid-size AI companies and downstream deployers are not directly in scope, but may be pulled in via contracts, partnerships, or disclosure expectations from their frontier-model vendors.
Does the December 2025 executive order preempt state AI laws?
Not by itself. The December 11, 2025 executive order establishes a DOJ AI Litigation Task Force and directs federal agencies to act, but courts must ultimately decide whether specific state AI laws are preempted. Until a ruling or new federal law, Colorado, California, and other state AI statutes remain in force.
Sources & Further Reading
- King & Spalding — New State AI Laws Effective January 1, 2026
- Baker Botts — Colorado AI Act Implementation Delayed
- Morgan Lewis — AI Enforcement Accelerates as Federal Policy Stalls and States Step In
- Sidley Austin — Unpacking the December 11, 2025 Executive Order
- WilmerHale — Transparency in Frontier Artificial Intelligence Act (SB 53)
