📋 A State-by-State Regulatory Patchwork Emerges

California Governor Gavin Newsom's September 2024 veto of SB 1047, the landmark AI safety bill authored by Senator Scott Wiener, was supposed to buy time for a federal framework. Instead, it triggered a cascade of state-level action. As of May 2026, five states have enacted comprehensive AI safety laws: Colorado's AI Act (effective July 1, 2026), Connecticut's SB 2 on algorithmic discrimination, the Texas AI Safety Act (TASA), New York's Automated Employment Decision Tools Act, and Washington's AI Transparency in Government Act.

Each addresses different aspects of AI governance, and none are fully compatible with the others.

"We're building the Tower of Babel of AI regulation," said Ryan Calo, professor at the University of Washington School of Law. "A model deployed nationally needs to comply with five different risk assessment frameworks, three different transparency reporting standards, and two different liability regimes. That's not just compliance cost—it's compliance chaos." The Software Alliance (BSA) estimates that multi-state compliance will cost the industry $4.2 billion annually, with 60% of that cost falling on startups and SMEs without dedicated regulatory teams.

📋 The Texas Approach: Kill Switches and Frontier Lab Accountability

The Texas AI Safety Act, signed by Governor Greg Abbott in June 2025 and taking effect January 2027, is the most prescriptive state law. It requires developers of frontier AI models (defined as those trained with over 10^26 FLOPs) to implement mandatory kill-switch mechanisms, submit safety test results to the Texas Attorney General, and carry liability insurance for harms caused by their models. The law explicitly creates a private right of action, allowing Texas residents to sue AI developers for damages—a provision that legal scholars note could generate significant litigation.

OpenAI, Anthropic, and Google have all submitted comments warning that the Texas law's threshold captures their current and near-future models, and that the kill-switch requirement is technically undefined. "What does a kill switch mean for an API that thousands of customers have integrated into production systems?" asked Anthropic's public policy director in testimony to the Texas House. "An abrupt takedown could cause more harm than the risks it's meant to prevent."

🏢 Congressional Deadlock and Industry Pivot to Preemption

Congress has held 14 hearings on AI regulation since the SB 1047 veto—the Senate Judiciary Committee's AI subcommittee alone has held eight—but no bill has reached a floor vote in either chamber. The Bipartisan Senate AI Working Group led by Majority Leader Schumer has produced three framework documents but no legislative text. The House AI Task Force disbanded in December 2025 without issuing recommendations.

In response, the industry coalition that once opposed SB 1047—including OpenAI, Microsoft, Google, and Meta—has reversed position and is now lobbying aggressively for comprehensive federal legislation with explicit preemption of state laws. The goal is to replace the emerging state patchwork with a single federal standard, ideally one less restrictive than the most stringent state laws. But the politics of AI preemption are challenging: state attorneys general, civil rights organizations, and labor unions all oppose losing state-level enforcement authority, creating an unlikely alliance between progressive groups and small-government conservatives who see state-level regulation as more accountable than federal bureaucracies.