AI Law Graveyard, Edition 1: 8 AI Bills That Died in 2024–2026 (And the Lessons Buried With Them)
Bill-tracker dashboards count dead bills as live coverage. We don't. Eight vetoed, enjoined, and quietly-killed AI bills — what they would have done, who killed them, and what filled the void.
Bill-tracker dashboards love to brag about how many AI bills they 'cover.' The honest answer is that most of those bills are dead. They were vetoed, gutted on a suspense file, pulled from markup, or blown up by a federal judge — and the trackers leave the corpses in the count without telling you. This column does the opposite. It treats the graveyard as the most useful part of AI policy: the place where you learn what governors will actually sign, what courts will tolerate, and what gap the next bill is going to try to fill. Every entry below gets the same four-part treatment: what it would have done, who killed it, what fills the void today, and what to watch for next. See the full archive of dead bills at /topics.
-
VETOED
1. California SB 1047 — Safe and Secure Innovation for Frontier AI Models Act
California (state) · Vetoed by Gov. Newsom, Sept. 29, 2024Sen. Wiener's bill would have imposed safety-testing, shutdown-capability, third-party-audit, and incident-reporting duties on developers of 'covered models' trained above $100M cost or 10^26 FLOPS, with whistleblower protections and AG enforcement. Newsom's veto message argued it would burden smaller models without targeting genuine frontier risks — language now recycled in every industry lobbying memo against frontier-AI bills. Nothing equivalent fills the void today. Watch for a revived SB 1047 with a higher FLOPS floor and narrower covered-model definition.
-
VETOED
2. Virginia HB 2094 — High-Risk AI Developer and Deployer Act
Virginia (state) · Vetoed by Gov. Youngkin, March 24, 2025Would have imposed Colorado-style duties on developers and deployers of high-risk AI systems — reasonable care to prevent algorithmic discrimination, impact assessments, consumer notice, incident reporting. AG enforcement, no private right of action. Youngkin's veto was the first red-state rejection of an EU-AI-Act-style framework, citing innovation concerns and Brussels overreach. Virginia now has no comprehensive AI deployer law. Watch for a pared-back 2027 bill targeting only employment and housing.
-
COURT
3. California AB 2839 — Materially Deceptive Election Content
California (state) · Preliminarily enjoined Oct. 2, 2024; 9th Cir. appeal fully briefed Mar. 11, 2026Criminalized materially deceptive AI-generated audio/video depicting candidates, election officials, or election processes within 120 days before or 60 days after an election. The Eastern District of California (Mendez, J.) preliminarily enjoined enforcement in Kohls v. Bonta on First Amendment grounds. Existing California defamation and election-tampering law now do the job, badly. If the 9th Circuit reverses, expect copycats in WA and OR; if it affirms, election-AI criminalization at the state level is effectively done for a generation.
-
COURT
4. California AB 2655 — Defending Democracy from Deepfake Deception Act
California (state) · Permanently enjoined Aug. 5, 2025 in Babylon Bee LLC v. Bonta (E.D. Cal.)Required large online platforms to remove or label materially deceptive AI election content within specified windows; created a private right of action — the only state platform deepfake statute with one. Permanently enjoined on Section 230 preemption and First Amendment grounds — the first permanent injunction of a state platform deepfake law on those theories, and likely the template defense in every future challenge. Nothing fills the void; platforms remain governed by their own moderation policies. Watch for legislatures pivoting from platform-targeted statutes to creator-targeted ones, which avoid Section 230.
-
VETOED
5. Federal — OBBBA Section 43201 (10-Year State AI Law Moratorium)
United States (federal) · Stripped 99-1 on the Senate floor, July 1, 2025Section 43201 of the One Big Beautiful Bill Act (H.R. 1, 2025) would have preempted all state laws limiting or imposing duties on AI models, systems, or automated decision systems for ten years. The Senate stripped it via a Blackburn amendment on a 99-1 vote — the lone dissent is contested across reports and we do not name it here. CA CCPA ADMT regs, CO SB 26-189, IL HB 3773, TX TRAIGA, and the NYC GUARD Act all exist because §43201 died. A second preemption attempt already failed in the FY26 NDAA. Watch the Obernolte / Trahan 'Great American AI Act of 2026' draft for narrower preemption language.
-
EXPIRED
6. Connecticut SB 2 — Artificial Intelligence Act
Connecticut (state) · Passed Senate Apr. 24, 2024; never called for House vote (session ended May 8, 2024)Sen. Maroney's comprehensive AI bill mirrored Colorado SB24-205 — duties for developers and deployers of consequential automated decision systems, a state AI Office, deepfake and GenAI-training disclosure requirements. House leadership never called it for a vote after Gov. Lamont opposition. A quiet death after months of public investment. Connecticut still has no comprehensive AI statute, but Maroney revived the framework as PA 26-15 (CART Act, June 2026). Watch whether CART Act enforcement clarifies what the 2024 opposition was actually about.
-
EXPIRED
7. California AB 3211 — Provenance, Authenticity, and Watermarking Standards
California (state) · Held on Senate Appropriations suspense file, Aug. 31, 2024Asm. Wicks' bill would have required covered GenAI providers to apply C2PA-style content provenance to AI-generated content, device manufacturers to enable provenance reading, and labeling of AI-detection tools — the most technically detailed provenance proposal any U.S. legislature has taken seriously. Killed on suspense, the procedural killshot that requires no recorded vote and produces no public reasoning. C2PA continues as a voluntary industry standard, with no statutory backing in U.S. law. Watch for a 2027 sequel scoped to political ads and synthetic media of identifiable persons.
-
EXPIRED
8. Federal — APRA AI / ADMT Provisions
United States (federal) · Pulled from House Energy & Commerce markup, June 27, 2024; never formally introducedThe American Privacy Rights Act discussion draft (Rodgers / Cantwell, April 2024) included covered-algorithm impact assessment requirements, opt-out rights for consequential automated decisions, and pre-deployment evaluation duties — the federal floor for ADMT rights that never was. Pulled from markup quietly: no vote, no opposition memo. CCPA ADMT regs, CO SB 26-189, and a patchwork of state biometric and employment laws now fill the floor APRA would have set. The lack of a federal ADMT framework is the single biggest reason the OBBBA preemption fight matters. Watch the Obernolte / Trahan draft and Sen. Cassidy's separate ADMT bill — APRA's ghost will be visible in the bill text.
What's filling the void: California's CCPA / CPRA ADMT regulations are the most consequential ADMT rules in the country, and they exist because the legislature couldn't pass SB 1001 / AB 331 / AB 3211 in clean form. Illinois' HB 3773 plus BIPA cover cross-sector AI discrimination and biometric harms. Texas TRAIGA is narrower than Colorado's law but it survived industry pressure. Colorado's rewritten SB 26-189 takes effect Jan. 1, 2027 and remains the closest thing to a U.S. EU-AI-Act analogue. State AGs are picking up the rest — see Texas AG Paxton's Assurance of Voluntary Compliance with Pieces Technologies (Sept. 2024), which did the work AB 3211's labeling rules would have done in California.
What to watch over the next 90 days: the 9th Circuit ruling in Kohls v. Bonta; the Obernolte / Trahan 'Great American AI Act of 2026' draft text; session-end deadlines on PA ADS and MI employment-AI bills that look structurally similar to 2024 ghosts; FY27 NDAA preemption language; CT PA 26-15 (CART Act) enforcement clarifications this fall.
The AI Law Graveyard publishes every other Friday. Full archive of vetoed, repealed, expired, and court-blocked AI laws at /topics; what's still alive is at the Bill Tracker.