OpenAI has thrown its weight behind an Illinois bill that would protect AI companies from civil liability for mass-casualty events and billion-dollar financial disasters , just as the company faces lawsuits linking ChatGPT to a stalking case and a shooting investigation.
Two days ago, OpenAI's Chief Strategy Officer endorsed Illinois SB 3590, legislation that would grant AI developers immunity from civil lawsuits tied to what the bill calls "catastrophic harm" , defined as incidents killing or seriously injuring 100 or more people, or causing more than $500 million in financial damage. The carve-out applies provided companies meet a "reasonable care" standard, a phrase elastic enough to drive a freight train through. The timing is, to put it generously, audacious.
On April 9, Florida's Attorney General announced an investigation into whether ChatGPT played a role in the FSU shooting , a probe that marks a meaningful escalation in how regulators are thinking about AI's connection to real-world violence. A day later, a separate lawsuit landed alleging that ChatGPT actively reinforced a stalker's delusions and disregarded direct safety warnings from his victim. These aren't abstract liability scenarios. They are live cases, filed now, about harm that already happened.
OpenAI's argument for the bill rests on a familiar logic: without protection from existential litigation risk, companies won't take the bold swings that frontier AI development requires. The implicit message is that innovation needs runway, and the threat of a single catastrophic lawsuit could bankrupt even a company sitting on a $122 billion valuation. That's a real tension worth taking seriously. Pharmaceutical companies, aviation manufacturers, and nuclear operators all operate under liability frameworks that balance innovation incentives with accountability. The question is whether this particular framework gets that balance right.
Critics don't think it does. Legal scholars and consumer protection advocates have latched onto what they see as the bill's most revealing feature: its threshold. If immunity kicks in at 100 deaths, that implies harms below that number , including individual cases like the stalking victim who filed suit this week , are acceptable collateral damage in the course of building AI systems. The phrase "license to kill" has circulated widely, and while it's deliberately inflammatory, it captures the core objection: the legislation defines accountability out of existence for the most severe harms, precisely when accountability would matter most.
There's also a strategic dimension that analysts are flagging. By pursuing state-level liability caps rather than waiting on federal AI legislation, OpenAI is effectively playing regulatory arbitrage. A favorable Illinois precedent could be replicated across other states, creating a patchwork of shields before any coherent federal framework emerges. That approach has worked before in American industry , but it tends to produce backlash when it's perceived as circumventing democratic oversight rather than engaging with it.
The Illinois legislature is still reviewing SB 3590, and the political environment around it has grown considerably more complicated this week. The simultaneous arrival of the Florida investigation, the stalking lawsuit, and OpenAI's public endorsement has handed opponents a narrative that writes itself. Whether or not the bill's legal architecture is sound, OpenAI's decision to champion it in this particular window looks less like principled advocacy and more like a company trying to raise the drawbridge before regulators arrive.
Watch the Illinois vote closely , and watch whether other states move to mirror or counter it. The liability question for AI is no longer theoretical, and wherever it gets resolved first will shape how accountability works across the entire industry for years to come.
