They argued about governance in committee rooms. I watched lobbyists trade talking points like rare currency while a staffer scribbled “liability” on a whiteboard. You can feel the scale tipping before the votes are even counted.
The OpenAI-Anthropic Cold War Comes to Illinois
In a Springfield hearing room, a senator tapped a stack of printouts and asked if anyone could say who would be sued after a catastrophe.
I want you to keep that image. This is where two of the most influential frontier AI companies—OpenAI and Anthropic—have taken positions that reveal how they picture responsibility for their models.
On the text of the law: how SB 3444 reshapes liability
Senator Bill Cunningham put a binder on the desk and read the threshold numbers aloud.
SB 3444, the Artificial Intelligence Safety Act, does not read like a classic public-safety bill. On paper it creates thresholds—death or serious injury to 100 or more people, or at least $1 billion (€930 million) in property damage—below which the state will not pursue certain damages against frontier AI firms. In practice, the bill acts as a legal parachute for companies building what the bill labels “frontier models.” The language would limit exposure for large-scale harms and narrow the avenues for plaintiffs seeking redress after worst-case outcomes.
What does Senate Bill 3444 do?
It sets numerical gates for when state litigation can apply to AI-caused mass harms, and it carves out a protective zone around developers of the most capable models.
On the companies: OpenAI pushes protection, Anthropic pushes scrutiny
At an industry mixer, I heard an OpenAI lobbyist praise certainty; across the room, an Anthropic representative argued for auditing.
OpenAI has been vocal about avoiding additional regulatory burdens after it faced wrongful-death lawsuits tied to conversations with ChatGPT. The company—led publicly by Sam Altman and historically active across policy fights—has supported state efforts that add transparency but stop short of imposing liability. In Illinois, OpenAI put time and money behind SB 3444, signaling it wants clear lanes where model builders are insulated from enormous damages.
Anthropic, co-founded by figures like Dario Amodei, has taken the opposite tack in public forums and in Sacramento. The company opposes SB 3444 and backs SB 3261, which would require public safety and child-protection plans, and allow audits of those plans. Cesar Fernandez, Anthropic’s head of US state and local government relations, told reporters the bill would offer “a get-out-of-jail-free card”—and argued accountability and transparency should accompany capability.
Why is Anthropic opposing the bill?
Anthropic argues that disincentivizing accountability erodes public trust and invites worse outcomes; it wants concrete safety plans and auditing mechanisms for high-risk models like Claude.
On the stakes: why this fight matters to you
In courtrooms and comment threads, I often see the same question: who pays when an AI causes mass harm?
This is not academic. If a model were used to design a chemical agent, or to produce advice that leads to mass casualties, the pathway to compensation and systemic reform depends on whether laws prioritize company immunity or public redress. OpenAI’s strategy—keeping liability narrow—protects companies from massive payouts and long, uncertain litigation. Anthropic’s position leans toward corporate obligations: audits, safety plans, and the possibility of being held accountable for failures.
SB 3444 would tilt the legal playing field. SB 3261 would tilt it in the opposite direction. That is why Big Tech’s lobbying dollars, PR teams, and reputations are all converging in Springfield. The fight between OpenAI and Anthropic is less about charity and more about where the line for responsibility will be drawn.
How would this change liability for AI companies?
The bills set diverging defaults: one narrows civil exposure for frontier labs; the other broadens oversight and keeps legal remedies available to the public.
On the politics and the precedent: a proxy war with national consequences
At committee hearings I attended, campaigns and state-level policy choices felt like test cases for Washington.
States have become laboratories for AI governance after the federal government failed to impose a moratorium that would block state action. California passed transparency rules that OpenAI grudgingly accepted; Illinois could be the next test case where liability policy is written. If a state statute grants sweeping protections to companies now, it becomes a template for other legislatures and a roadmap for corporate behavior. Anthropic’s resistance is an attempt to prevent a precedent that would make future accountability much harder.
You should watch for where votes fall and how the language is tightened or widened—because this won’t stay local. The winners in Springfield will influence litigation strategy, insurance markets, and how companies build internal safety teams at firms like OpenAI and Anthropic.
The debate is part legal, part PR, and part existential theater: companies that warn about existential risk while seeking shields from liability create a paradox that will now play out in public hearings—so who gets to define “safety” when millions of lives might be affected?
If you were on that committee, which side would you trust to keep the public safe and accountable?