President Newsom: The First 100 Days of Federal AI Regulation

Hero photo: Gavin Newsom official photo — Office of the Lieutenant Governor of California, Wikimedia Commons, public domain.

01 The record and the scenario

Scenario analysis, not a prediction or endorsement: as of September 18, 2026, AP describes Gavin Newsom as a potential 2028 Democratic contender among a field where most figures profiled have not formally entered a presidential race. Nobody wins anything from this exercise. The question is narrower: if Gavin Newsom were sworn in on January 20, 2029, what would the first 100 days of federal AI policy actually look like?

His record is unusually legible. In September 2023 he signed Executive Order N-12-23, ordering state agencies to study generative AI and to write procurement rules for its use in California government — no binding industry mandates. In September 2024 he vetoed SB 1047, Scott Wiener's frontier-AI safety bill, writing that the measure could impose burdens on smaller developers and give California an outsized role before the technology had matured — while pledging to work on a more balanced approach. In September 2025 he signed SB 53, the Transparency in Frontier Artificial Intelligence Act: first-in-nation requirements for frontier developers to disclose safety plans, protections for AI whistleblowers, and a public compute cluster (CalCompute). And in September 2026, with Washington having produced no federal framework, he signed an executive order advancing frontier-AI oversight — directing state agencies to consult experts on possible kill-switch requirements for the largest systems and calling on the federal government to act.

That arc — voluntary first, veto the rigid version, sign the transparency version, then prepare binding standards — is the closest thing to a Newsom doctrine on AI. A President Newsom would arrive with a finished, tested playbook and a claim no other contender can make: he has already governed the industry's home state, and it still votes for him.

02 Day 1: executive orders

Day 1 is where a Newsom presidency would look most different from both a second Trump-term AI posture and the Biden approach it replaced. The instrument is the same one he used in Sacramento: the executive order. The available moves under existing authority:

Repeal and replace the AI executive-order stack. Recent administrations have alternated — the Biden EO 14110 on Safe, Secure, and Trustworthy AI was rescinded in January 2025 in favor of a deregulation order; a Newsom EO would chart a third path, restoring the federal inventory of AI uses and risk-classification work EO 14110 began while rejecting both the recision and the rigid licensing ideas he vetoed in SB 1047.

Direct OSTP and the AI safety institutes. The U.S. AI Safety Institute at NIST, created under the Commerce Department, would get an explicit mandate to draft frontier-model safety-reporting standards — the federal analog of what California required by statute in SB 53. The National Security Memorandum on AI from 2024 already treats frontier AI as a national-security asset; Newsom would keep that framing and add reporting teeth.

Procurement as quiet regulation. The federal government buys enough software that its contract terms become de facto standards. A Day-1 order instructing the General Services Administration to require AI safety plans and incident-reporting from vendors supplying frontier systems would move the whole industry without a single vote in Congress.

03 Days 2-30: agency changes

The first month would be personnel and memo work — the part of an AI agenda that gets least coverage and matters most. The pattern from Sacramento: a governor who staffs agencies with industry insiders and consumer advocates simultaneously, and forces them to draft rules together.

The moves available in this window: a directive to the Office of Management and Budget updating the federal AI procurement and use guidance (the OMB memos that implement executive orders do the real work); a Federal Trade Commission already empowered to police deceptive AI claims, with new leadership prioritizing model-performance fraud; and NIST instructions to expand the AI Risk Management Framework toward an auditable reporting template — a voluntary document that becomes binding the moment a procurement rule incorporates it by reference.

The veto message matters for what it forecloses. Newsom explicitly declined the SB 1047 approach — pre-deployment safety certifications for large models. A Newsom FTC or NIST would push disclosure and liability for false claims, not licensing gates. The industry knows the difference: he is the Democrat who killed the bill Silicon Valley feared most, and the one who then signed the law it could live with.

04 The first budget

The first budget request is where the priorities are visible. In Sacramento, the governor's January budgets funded what EO N-12-23 started — a state GenAI training series for civil servants, dedicated agencies for AI incidents, and eventually CalCompute, the state-backed compute cluster in SB 53. Those are exactly the line items a first Newsom budget would carry to Washington:

AI Safety Institute funding. AISI at NIST has operated on thin, uncertain appropriations; a Newsom budget would make it a standing program with statutory-scale funding, on the theory that the U.S. needs a government lab capable of actually testing frontier models the way the FAA tests aviation systems.

Public compute. CalCompute is the tell. Newsom has argued that safety research is captured when only the largest companies can afford the compute to evaluate their own models. A federal equivalent — research compute available to universities and civil society — would be a new budget line with a built-in constituency.

State-government AI grants. The California playbook of funding state agencies to safely adopt AI, scaled through federal grants-in-aid, would let a President Newsom build a national record without waiting for Congress to pass a framework statute.

05 First legislation and what requires Congress

Here the constraint is structural: everything binding on private AI developers requires Congress. Executive orders bind agencies; statutes bind companies. A President Newsom would face the same Congress that has considered — and not passed — every major AI bill of this decade, including the bipartisan AI framework proposals senators negotiated in 2024 and again in 2025 without reaching the floor.

What could actually move: a transparency-first bill modeled on SB 53 — safety-plan disclosure, whistleblower protection, incident reporting — is the piece with the broadest coalition, since it borrows from aviation and cybersecurity precedent rather than creating a new regulator. It is the same bill Newsom signed with tech companies at the signing table in Sacramento.

What cannot move in 100 days: pre-deployment licensing (the idea he vetoed), comprehensive AI liability, and any framework preempting the state laws — including his own SB 53 — that currently fill the federal vacuum. Reconciliation, the budget procedure that bypasses the Senate filibuster, could fund institutes and compute but cannot write regulatory law, because reconciliation rules bar provisions that are merely policy, not spending or revenue. The first 100 days deliver a rulemaking agenda; the statute comes later or not at all.

06 Foreign policy: the AI competition frame

AI policy is foreign policy. The 2024 National Security Memorandum on AI and the diffusion framework the Commerce Department uses to restrict advanced-chip exports treat frontier models as strategic assets; any administration inherits a contest with China that shapes what regulation is politically possible. Newsom's positioning — a Californian who regulates the industry without driving it offshore — is itself a China argument: American frontier labs stay competitive, but under enforceable disclosure rules that Beijing's system cannot demand of its own champions.

The concrete first-100-days moves are diplomatic and export-control adjacent: coordinating safety-reporting standards with the UK and Japanese AI safety institutes that already work with the U.S. AISI; pressing at the G7 to align frontier-model disclosure norms before the EU AI Act's phased obligations become the world's default; and holding the export-control line on advanced accelerators — the one lever a president controls outright. The international story of a Newsom administration would be an attempt to build an American-led transparency bloc before anyone else's model hardens.

07 What courts could constrain

Every tool in the 100-day kit is judicially contestable. Executive orders ride on delegated authority: a 2025 Supreme Court decision held that federal courts, not agencies, decide the meaning of ambiguous statutes — which means an ambitious AI EO that reads new powers into existing law invites challenges across the doctrine's open ground. Procurement-based regulation is safer legally but vulnerable to the same major-questions doctrine that has clipped agency reach when agencies assert transformative authority without clear congressional direction.

The First Amendment is the AI-specific frontier. Courts have already enjoined state laws restricting algorithmic outputs — a federal judge blocked Texas's age-verification law for censoring speech, and the same logic would test any federal mandate that model outputs be filtered or disclosed in specific ways. A kill-switch requirement for the largest models, translated from the 2026 California EO into a federal rule, would be litigated the day it is proposed.