AI Regulation Fight Deepens Over Liability and Power

The fight over artificial intelligence regulation has entered a sharper, more consequential phase: it is no longer simply a contest between the tech industry and government, but a widening split among the leading AI labs themselves over who should be held responsible when powerful systems cause catastrophic harm.

A Fault Line Over Liability

In Illinois, this divide has become unusually explicit. A state bill backed by OpenAI would narrow the circumstances in which AI developers could be held liable when their models are used to inflict extraordinary damage, including mass casualties or more than $1 billion in property losses. Anthropic, another leading AI company often grouped with OpenAI as part of the industry’s safety-conscious wing, has been lobbying lawmakers to block or significantly rewrite the measure.

The dispute centers on Senate Bill 3444, introduced by State Senator Bill Cunningham. The proposal would limit the legal exposure of AI labs when their models are used to produce extreme harms. OpenAI has supported the measure as a way to focus attention on the most serious risks while avoiding a confusing patchwork of state standards. This reflects the company’s broader position: that AI regulation should be uniform and federal, rather than a collection of varying state rules.

Anthropic has broken sharply with that approach. Rather than backing the effort to cabin liability, it has reportedly pushed Illinois lawmakers toward tougher alternatives that would preserve or expand safety obligations for developers. This disagreement is notable because it turns on a fundamental question: When harm occurs, who pays?

From Statehouses to Campaign Warfare

The political battle is escalating beyond committee rooms and into the electoral arena. In New York, Alex Bores, a state assemblyman and former Palantir employee, has become a target of a deep-pocketed super PAC backed by major Silicon Valley figures.

Bores helped advance the RAISE Act, a frontier-model safety law signed by Governor Kathy Hochul. The law established a state framework for transparency, safety planning, and incident reporting. Now, Bores is being targeted by “Leading the Future,” a pro-AI super PAC tied to powerful donors including OpenAI president Greg Brockman and Andreessen Horowitz.

This campaign marks a significant escalation. Because Washington has moved slowly, states like California and New York have become laboratories for frontier-AI policy. If the industry begins treating state legislators as political threats to be neutralized, it could chill state experimentation before federal regulators ever step in.

A Deeper Split Among AI Labs

The clash undercuts the common framing of the AI debate as a simple showdown between “innovation” and “restraint.”

Anthropic and OpenAI have signaled different instincts on state-level oversight before. In California, Anthropic supported a transparency-focused framework, while OpenAI argued against state-by-state requirements. The Illinois dispute carries that divergence into the more combustible terrain of tort law and catastrophe liability.

The OpenAI Position: Developers should not face boundless liability for how third parties misuse general-purpose systems, especially since the software can be adapted in ways outside a company’s control.
The Critic/Anthropic Position: Overly broad liability shields reduce incentives for labs to build strong safeguards or conduct rigorous pre-deployment testing.

Why the Battle Matters Now

The timing is critical. As AI systems become capable of writing code, generating synthetic media, and potentially assisting in the design of dangerous biological agents, the juridical boundary between a “toolmaker” and a “malicious user” grows harder to draw.

A law enacted in Illinois or New York rarely stays local; it often becomes a de facto national compliance standard. The rhetorical consensus that AI “should be regulated” is officially over. The new reality is a concrete struggle over exactly where accountability lands—and who gets to decide.

Sources

Further reading and reporting used to add context:

AI News