The December 2025 Executive Order on preempting state AI laws, mapped to SWT3 witness procedures that produce jurisdiction-portable compliance evidence.
Who this is for: General counsel, compliance officers managing multi-state AI deployments, government affairs teams tracking federal preemption developments, and AI policy analysts advising organizations on regulatory strategy.
Current status (September 2026): This Executive Order is policy direction, not legislation. It instructs federal agencies to identify and challenge state AI laws, but actual preemption of state law requires Congressional action or federal court rulings. All referenced state laws remain enforceable until Congress acts or courts rule otherwise. Organizations should continue complying with applicable state requirements.
On December 11, 2025, the White House issued an Executive Order titled "Ensuring a National Policy Framework for Artificial Intelligence," also referred to as "Eliminating State Law Obstruction of National Artificial Intelligence Policy." The EO establishes a federal policy position that state-level AI regulation creates an inconsistent patchwork that burdens interstate commerce and innovation.
Even the EO's own framework carves out categories of state regulation that would survive federal preemption:
The following table summarizes major state AI laws and their exposure to federal preemption under the EO's framework. "Preemption risk" reflects the law's likelihood of being targeted based on the EO's stated criteria: AI-specific regulation, burden on interstate commerce, and inconsistency with federal deregulatory policy.
| State | Law | Core Requirement | Effective | Preemption Risk |
|---|---|---|---|---|
| Colorado | SB 26-189 | ADMT transparency, pre-use notice, human review on request | Jan 1, 2027 | High |
| Illinois | HB 3773 | AI in employment decisions, bias audits, notice requirements | Jan 1, 2026 | High |
| Connecticut | SB 2 (CART Act) | Impact assessments for high-risk AI, consumer opt-out rights | Oct 1, 2027 | High |
| Utah | HB 276 | AI interaction disclosure, regulatory sandbox | May 1, 2024 | Medium |
| Texas | TRAIGA | Inventory of state AI use, transparency for government AI | Sep 1, 2025 | Low |
| California | AB 2013 | GenAI training data transparency | Jan 1, 2026 | Medium |
| California | SB 942 | AI-generated content disclosure and detection tools | Jan 1, 2026 | Medium |
| New York | LL 144 | AEDT bias audits in hiring | Jul 5, 2023 | High |
| Maryland | HB 895 | Facial recognition restrictions in hiring | Oct 1, 2024 | Medium |
| Tennessee | SB 1580 | ELVIS Act, voice/likeness AI protections | Jul 1, 2024 | Low |
Risk assessment criteria: Laws rated "High" are AI-specific regulations imposing affirmative compliance obligations on private-sector deployers across state lines. Laws rated "Medium" apply to specific use cases or have narrower scope. Laws rated "Low" focus on government use of AI or address intellectual property rather than AI governance.
Important: "High preemption risk" does not mean the law will be preempted. It means the law matches the profile the EO targets. Congressional action is required, and any preemption legislation will face significant political debate. As of September 2026, no preemption bill has passed either chamber.
The preemption debate creates genuine uncertainty for compliance teams. Building evidence exclusively for one state law risks wasted investment if that law is preempted. Ignoring state laws risks enforcement if preemption fails or takes years.
SWT3 witness procedures resolve this tension because they record governance-relevant events at a level of abstraction that is independent of any specific jurisdiction's requirements.
Regardless of whether preemption succeeds, the EO and its March 2026 framework articulate expectations that any federal AI law will likely include. The following table maps those expectations to SWT3 procedures.
| Federal Expectation | Source | SWT3 Procedure | Why It Survives Preemption |
|---|---|---|---|
| Transparency / disclosure | EO Sec. 3(b), March Framework Sec. 4 | AI-TRANS.1 | Every proposed federal framework includes disclosure obligations. Transparency is bipartisan. |
| Explainability | March Framework Sec. 5 | AI-EXPL.1 | Fundamental to both state and federal approaches. Required independently under ECOA, FCRA. |
| Bias measurement | EO Sec. 4(a), Civil Rights Act | AI-FAIR.1 | Civil rights obligations exist independently of AI-specific law. Title VII, ECOA, FHA all apply. |
| Audit trail / recordkeeping | March Framework Sec. 7 | AI-AUDIT.1 | Recordkeeping survives any regulatory framework. Every compliance regime requires demonstrable evidence. |
| AI governance documentation | EO Sec. 5, NIST AI RMF | AI-GOV.1 | Governance documentation is useful regardless of which laws apply. NIST AI RMF is non-preemptable (voluntary). |
| Consent and notice | March Framework Sec. 4 | AI-CONSENT.1 | Consumer protection basics survive preemption. FTC Act Section 5 applies independently. |
| Human oversight | EO Sec. 3(c), March Framework Sec. 6 | AI-HITL.1 | Human oversight is expected in every proposed federal framework. Likely required for high-risk decisions. |
| Automated decision notification | March Framework Sec. 4 | AI-AUTO.1 | Notification of automated decisions is a bipartisan expectation. Required under FCRA adverse action rules. |
| Data provenance | March Framework Sec. 8 | AI-DATA.1 | Data governance is framework-neutral. Training data documentation expected under any regime. |
| Incident response | EO Sec. 6, NIST AI 600-1 | AI-INCIDENT.1 | Incident reporting is expected in any regulatory framework, federal or state. |
| Inference provenance | NIST AI RMF, March Framework Sec. 7 | AI-INF.1 | Runtime evidence transcends specific regulations. Model, prompt, and response hashing is fundamental. |
Federal expectation: Both the EO and March 2026 framework call for "clear and accessible" disclosure when AI is used in decisions affecting individuals. This mirrors requirements in Colorado SB 26-189, Illinois HB 3773, and the Connecticut CART Act.
How SWT3 addresses it: witnessTransparency() mints an anchor recording the disclosure type (AI usage notification, system capability description), recipient type (end user, data subject, regulator), and delivery timestamp. The anchor is jurisdiction-neutral, proving that disclosure occurred regardless of which law requires it.
Preemption analysis: Transparency disclosure is present in every proposed federal AI framework, the EO's own March 2026 document, and all major state AI laws. This obligation will exist whether federal or state law governs.
Query the witness ledger for AI-TRANS.1 anchors filtered by deployment jurisdiction. Each anchor's timestamp proves when disclosure was generated. The same evidence set satisfies Colorado's pre-use notice, Illinois's notification requirement, Connecticut's disclosure obligation, and any future federal transparency mandate.
Federal expectation: The EO acknowledges that AI systems must not violate civil rights protections. The March 2026 framework references NIST AI RMF's fairness functions. Existing federal law (Title VII, ECOA, FHA) already prohibits discriminatory outcomes from automated systems.
How SWT3 addresses it: witnessBiasMeasurement() mints an anchor recording demographic group count, maximum disparity ratio, and measurement methodology. This evidence is valuable under state bias audit laws (NYC LL 144, Illinois HB 3773), existing federal anti-discrimination statutes, and any future federal AI fairness requirements.
Preemption analysis: Civil rights obligations are independent of AI-specific law. Even if every state AI law is preempted, Title VII, ECOA, and the FHA still apply to AI-driven decisions in employment, lending, and housing. Bias measurement evidence is always relevant.
AI-FAIR.1 anchors record disparity ratios per protected group. Under state law, these satisfy bias audit requirements directly. Under federal law, they demonstrate proactive disparate impact monitoring under Griggs v. Duke Power standards. The evidence is the same; the legal framework referencing it changes.
Federal expectation: The EO's Section 3(c) calls for "appropriate human oversight" of AI systems making consequential decisions. The March 2026 framework proposes "meaningful human review" requirements for high-risk AI applications, mirroring Colorado's consumer-request review right and the EU AI Act's Article 14.
How SWT3 addresses it: witnessHumanReview() mints an anchor recording the reviewer identity, review outcome, and whether the AI recommendation was affirmed or overridden. Combined with AI-HITL.2 (override tracking), this creates a complete chain proving that human oversight was substantive.
Preemption analysis: Human oversight requirements appear in every proposed federal AI framework. The EO itself advocates for human review of high-risk AI decisions. This obligation is preemption-proof because both sides of the debate agree on it.
AI-HITL.1 anchors prove reviews occurred. Under state law, they satisfy human review on request. Under federal frameworks, they demonstrate the "meaningful human oversight" that every proposal requires. Absence of AI-HITL.2 (override) anchors across large decision volumes may indicate rubber-stamping, which is a concern under any framework.
Federal expectation: The March 2026 framework includes recordkeeping requirements for AI systems making consequential decisions. NIST AI RMF (Govern 1.7, Map 3.5) requires documented governance processes. Existing sectoral regulations (FFIEC, OCC, FDIC) already mandate recordkeeping for automated decision systems.
How SWT3 addresses it: Every SWT3 Witness Anchor is written to the sovereign witness ledger with a cryptographic fingerprint. witnessAuditIntegrity() periodically verifies the log's tamper-evident properties. The ledger itself functions as a portable compliance record with independently verifiable entries.
Preemption analysis: Recordkeeping is the most durable compliance obligation. Every regulatory regime, federal or state, requires demonstrable evidence of compliance. An auditable trail of AI governance events is useful under any legal framework that could emerge.
Export the witness ledger for the relevant time period. Each anchor's SHA-256 fingerprint can be independently recomputed. Under Colorado's 3-year retention, the ledger is the compliance record. Under federal recordkeeping rules, the same ledger satisfies the same need. Daily Merkle rollups provide batch verification for any auditor.
Federal expectation: The March 2026 framework describes expectations for documenting "what model produced what output, on what input, at what time." NIST AI 600-1 (GenAI Profile) identifies provenance as a core governance function. This expectation exists independently of any AI-specific statute.
How SWT3 addresses it: witness() mints an anchor for every inference, recording the model identifier, prompt hash, response hash, and timestamp. This is the foundational SWT3 operation, creating a cryptographic record of every AI decision at runtime without impacting latency.
Preemption analysis: Inference provenance is a technical governance capability, not a compliance obligation tied to any specific law. It is equally valuable under state law, federal law, voluntary frameworks, and contractual requirements. Organizations building inference provenance today are investing in a capability that no regulatory outcome can render unnecessary.
AI-INF.1 is the most jurisdiction-neutral procedure in the SWT3 protocol. An inference anchor proves which model processed which input at which time. This evidence is useful for state transparency laws, federal audit requirements, contractual SLAs, incident forensics, and model drift detection. Ask the deployer for inference anchor density (anchors per decision) to assess coverage.
| Date | Event | Significance |
|---|---|---|
| Dec 11, 2025 | Executive Order issued | Establishes federal policy position against "burdensome" state AI regulation. Directs 90-day agency review. |
| Mar 11, 2026 | 90-day agency review deadline | Federal agencies submit lists of state laws identified as inconsistent with federal deregulatory policy. |
| Mar 20, 2026 | "National Policy Framework for AI" released | Follow-up document with legislative recommendations for Congressional action on preemption. |
| Apr 27, 2026 | Colorado SB 24-205 stayed | Original Colorado AI Act implementation halted pending replacement legislation. |
| May 14, 2026 | Colorado SB 26-189 signed | Replacement law with narrower ADMT transparency focus. Effective January 1, 2027. |
| Sep 2026 | Congressional status | No preemption bill has passed either chamber. Multiple proposals in committee. State laws remain enforceable. |
| Jan 1, 2027 | Colorado SB 26-189 effective | ADMT transparency requirements take effect unless preempted before this date (unlikely). |
| TBD | Congressional action | Preemption requires legislation. Timeline is uncertain and depends on political dynamics. |
Planning assumption: State AI laws will remain enforceable for at least 12-24 months beyond September 2026. Even if Congress passes preemption legislation, implementation timelines and legal challenges will extend the transition period. Organizations should plan for continued multi-state compliance.
The EO is a policy signal, not a legal safe harbor. Organizations that pause state AI compliance based on the EO assume significant risk. If preemption fails or is delayed (which is the most likely near-term scenario), those organizations will have compliance gaps with no evidence to close them.
Use SWT3 witness procedures to generate cryptographic evidence of AI governance events. This evidence is not tied to any specific state law. It serves as portable compliance infrastructure that maps to any current or future regulatory framework.
Focus on procedures that satisfy obligations across every proposed framework, both state and federal:
Track Congressional action on AI preemption legislation. Key committees: Senate Commerce, Science, and Transportation; House Energy and Commerce. Any enacted legislation will require a crosswalk update, but existing SWT3 evidence anchors will map to new requirements through the framework crosswalk engine.
The most likely long-term outcome is not total preemption or total state autonomy, but a cooperative model where federal law sets a floor and states retain authority in specific domains (child safety, procurement, civil rights). SWT3's procedure-level architecture is designed for exactly this scenario, where evidence must satisfy overlapping requirements from multiple jurisdictions simultaneously.
| General Counsel Question | Where to Look |
|---|---|
| Are we safe to stop state AI compliance? | No. The EO is policy direction, not law. State laws remain enforceable. Continue compliance and build portable evidence with SWT3 procedures. |
| Which state laws are most at risk? | AI-specific laws imposing affirmative compliance obligations on private-sector deployers: Colorado SB 26-189, Illinois HB 3773, Connecticut CART Act, NYC LL 144. See Section 2 table. |
| What evidence survives preemption? | Evidence of transparency (AI-TRANS.1), fairness (AI-FAIR.1), human oversight (AI-HITL.1), audit integrity (AI-AUDIT.1), and inference provenance (AI-INF.1). These map to both state and proposed federal requirements. |
| What if a federal AI law passes? | SWT3's framework crosswalk engine maps 286 procedures across 77 frameworks. Adding a new federal framework is a database operation. Existing witness anchors immediately map to new requirements. |
| How do we prove compliance to both state and federal auditors? | SWT3 clearing levels allow the same evidence to be shared at appropriate sensitivity levels. L0 for analytics, L1 for standard audit, L2 for sensitive detail, L3 for classified review. |
| What is the cost of inaction? | If preemption fails, organizations without evidence face enforcement with no compliance record. If preemption succeeds, organizations with evidence have a head start on federal requirements. The asymmetry favors action. |