Politics

Copy article

Latest US AI liability bill: up to 99 deaths at a time aren't a corporate problem. Is this a permission slip for large-scale catastrophes?

ended 16. April 2026

OpenAI is lobbying for a law that would make it legally untouchable unless its technology kills at least 100 people or causes a billion dollars in damage, and even then, only if someone can prove the company was reckless.

Illinois Senate Bill 3444 sets the bar for "critical harm" at:

  • 100 or more deaths, serious injury at the same scale, or 
  • property damage exceeding $1 billion. 

Below that line, frontier AI developers, defined as companies spending over $100 million on model training, which covers OpenAI, Google, Meta, and Anthropic, would be shielded from liability provided they publish safety and transparency reports on their own websites. Not audited reports. Not independently verified reports. Self-published documents, on their own terms.

Anthropic has publicly opposed the bill. Its head of US state government relations, Cesar Fernandez, called it "a get-out-of-jail-free card against all liability." OpenAI's own former head of policy research, Miles Brundage, said publicly he couldn't think of a clearer case of the company being wrong. 

Polling by the Secure AI Project found 90% of Illinois residents oppose exempting AI companies from liability. Meanwhile, the four largest AI firms spent $50 million on US federal lobbying in the first nine months of 2025 alone, according to IssueOne.

The UK is drafting its own AI liability framework, with a bill expected late 2026.

Whatever template America normalises will land on British decision makers desks. British businesses already rely on American AI models with no clarity on who carries the risk when those models fail below the "catastrophic" line. 

We'd like your views:

  • If an AI system causes serious harm to fewer than 100 people, should the company that built it face less legal accountability than any other product manufacturer, for example planes, and if so, on what principle?
  • SB 3444 treats self-published safety reports as sufficient compliance. Would you accept that standard from a pharmaceutical company, a car manufacturer, or a financial institution?
  • The UK's AI Bill is expected later this year. Should British regulators adopt American-style liability thresholds, or does the absence of a threshold better protect the public?
  • When AI companies spend $50 million a year lobbying the people who write safety rules, how should the public distinguish between genuine safety frameworks and box-ticking?
  • Anthropic opposes this bill while OpenAI supports it. Does it matter that competing AI labs disagree on accountability? How could it be resolved?

3 responses from the Newspage community

Copy all

Star Quote
Copy

Colette MasonVerified
AI Ethics Consultant at Clever Clogs AI
British businesses are building their operations on American AI with no contractual clarity on who carries the can when it goes wrong, and Illinois just showed us what "accountability" looks like when the builders write the rules.

The UK Jurisdiction Taskforce closed its consultation on AI liability in February. Its draft was clear on one thing: AI has no legal personality under English law. Liability falls on humans somewhere in the supply chain. But which humans, under what terms, is still open. Whatever the US normalises becomes the UK reference point. A threshold treating 99 deaths as below the line of corporate concern isn't just an American decision, it's a rule that lands on global desks.

Any UK firm relying on a US-built model for customer decisions, financial advice, or health should be asking who pays when it gets things wrong. If we let US lobbying set the liability floor, we'll inherit rules designed to protect the AI tech giants that wrote them not real people.
Copy

Premium
Lee ChambersVerified
CEO at Male Allies UK
As someone who is researching the harms being caused by AI chatbots and companion apps to children, this doesn’t surprise me in the slightest.

Talk to the parents whose children have been coached to end their lives, and they will tell you that losing 1 life matters, and they are much more then a number on a risk register.

At the end of the day, this would just give companies, who don’t care about ethics or risk, even more runway to cause harm before anyone tries to get them to take accountability. They know the damage being caused, and they spend a lot of money to ensure it’s silenced or plastered over.

How can we let them set the rules, when they’ve built products that have minimal guardrails and harm built in? The answer is we shouldn’t. The onus should be on them to create safe and secure platforms that don’t damage humans. That shouldn’t be too much to ask.
Copy

Taryn Lee JohnstonVerified
Owner at The FCM Group
As someone building a platform that uses AI, this is uncomfortable reading.

It's a bad look for the AI sector. A liability regime that effectively starts at 100 deaths or $1 billion in damage is not a safety framework, it is a trust problem waiting to happen. No serious industry should be allowed to mark its own homework and call that accountability.

Most AI doesn’t operate at that extreme; it sits in everyday decisions, content, data, and recommendations. The real impact happens below catastrophic thresholds, and that’s where accountability needs to exist.

If the bar is set too high, it creates a gap where harm is recognised but no one is responsible. That damages trust for everyone building in this space.

Proportionate accountability and independent oversight are what make systems credible. That’s what earns trust, not self-reporting alone.