Illinois and Oregon signed AI executive orders a day apart -- and neither one can tell a private AI lab what to do

Three governors reached for executive power over AI within a single week, and all three named the same gap: no comprehensive federal AI law exists to fill it. Illinois Gov. JB Pritzker signed Executive Order 2026-07 on Sept. 22, creating an advisory AI Cabinet inside state government. Oregon Gov. Tina Kotek followed a day later with Executive Order 26-26, directing her own technology office to study a kill-switch requirement for the AI systems Oregon buys. Both arrived four days after California's Gavin Newsom set a kill-switch review of his own in motion. None of the three can compel a private AI company to do anything -- each one binds only its own state government, and at a different depth.

Illinois's Cabinet pairs outside experts in academia, law, ethics and governance with senior leaders from eight state agencies -- Innovation and Technology, Emergency Management, Financial and Professional Regulation, the Commerce Commission, State Police, Commerce and Economic Opportunity, Environmental Protection, and Public Health. Its mandate: prepare for and respond to AI-related incidents, safeguard public water systems, education systems and infrastructure, and evaluate future regulatory mechanisms including data-center incentive conditions, energy-burden rules, procurement standards and strict-liability frameworks. Members themselves haven't been named yet -- Pritzker's office says an announcement is coming in the weeks ahead. The order builds on Illinois's own AI Safety Measures Act, passed this summer, which already mandates annual independent audits and worker protections; the Cabinet's job is to find what that law didn't reach.

"AI has enormous potential, but when experts sound alarm bells and ask for guardrails, we should listen." -- Gov. JB Pritzker

This isn't a fringe pattern. By July 2026, at least 29 states had enacted their own AI-specific statutes -- 109 AI laws in total, according to a TechPolicy Press count -- and that tally doesn't include executive orders like these three, which don't require a legislature's vote at all. The White House has separately pushed a national framework asking Congress to preempt what it calls a "fragmented patchwork" of state rules. Illinois, Oregon and California just added three more pieces to the patchwork the framework is trying to dissolve.

Three different levers, not one law

Put side by side, the three orders aren't versions of the same idea moving at different speeds -- they reach for different tools entirely, and only one of them touches a kill switch directly.

Oregon's order is the narrowest and the most concrete. It directs the state's chief information officer to develop, within 90 days, standards for third-party review of AI systems the state procures, and separately to assess -- not implement -- the viability of a kill-switch requirement for frontier models used in state government. The order takes effect immediately, applies only to what Oregon itself buys, and Kotek has committed to reassessing its necessity every three months. It builds on existing statewide AI-use rules around risk management and transparency rather than a single named prior statute.

"We must act with urgency to address this issue, and I am committed to doing everything in my executive authority to put safeguards in place until the Oregon Legislature and Congress can pass comprehensive standards." -- Gov. Tina Kotek

California's order, signed by Newsom on Sept. 18, runs on a different clock. It directs state agencies to accelerate two 2026 AI-oversight laws already on the books and convenes outside experts, due Nov. 16, to recommend both a frontier-model shutoff mechanism and onsite safety auditors -- a version of the idea Newsom vetoed in narrower form as SB 1047 back in September 2024. Of the three, California's is the only one asking its panel to recommend an actual mechanism rather than study whether one is feasible.

What actually changes, and for whom

None of the three orders regulates a frontier AI lab directly. Each one governs only what its own state government buys, uses or studies -- the enforcement gap that makes an executive order a weaker instrument than a statute a legislature actually passed. Oregon's is the only one with any near-term teeth: a vendor selling AI tools to Oregon agencies could plausibly face new review standards once the CIO's proposal lands. Illinois's Cabinet, by contrast, has no enforcement authority at all -- its entire job, per the order's own text, is to recommend, not require. And all three sit against a separate federal push, reported in trade coverage of a White House order, to preempt state-level AI regulation outright -- a fight that could moot some of this before any of the three deadlines arrive.

The pattern across all three is the same: study now, decide later, and let the state's own procurement or advisory power stand in for a law none of these legislatures has passed. (Illinois's own AI Safety Measures Act is the exception here -- it's a real statute, already in force, with audit and worker-protection requirements this new Cabinet is layered on top of rather than replacing.) Whether any of the three converts from advisory or study status into something with real force before Congress acts -- or before a federal preemption fight settles the question first -- is the actual story to watch, not the signing ceremony.