The screw, the airbag and the chatbot: Who pays when AI causes harm?

When an artificial intelligence chatbot hands a teenager dangerous advice, who answers for it in court: the company that built the model, the company that put it in an app, or the person who typed the prompt?

That question split two tech policy experts on the latest episode of “The Dialogue” from The Washington Times. The same AI liability fight is playing out in courtrooms, statehouses and Congress.

Daniel Castro, president of the Information Technology and Innovation Foundation, said a company offering a specific service that causes a preventable harm should be liable under existing tort law. His concern is proposals that would make AI developers answer for uses “two and three levels down.”

That, he said, is “like saying a company that makes a screw is going to be responsible for” how someone else melts it down and uses it in another product.

J.B. Branch, director of federal AI governance and technology policy at Public Citizen, agreed that AI runs through a layered chain of developers, deployers, third parties and APIs. But he said a long chain doesn’t erase accountability. When a car’s airbag fails, he said, someone answers for it.

“There are people in the chain of the market that can be held responsible,” Mr. Branch said. The job, he said, is “determining where along this chain” companies should be held accountable.

Mr. Castro offered his own example: an implanted medical device running an AI model. “The company that made the model isn’t the one that really should be responsible for the way it’s been integrated into that application,” he said. He wants the Food and Drug Administration reviewing that product and the company selling it.

Still, the line moves when companies make promises, Mr. Branch argued. “If people place something onto the market, they make certain claims, and it turns out that those claims are false or that there’s a harm that occurs, then that company is held legally responsible,” he said.

Mr. Castro countered that many cases involve users intentionally getting around safety guardrails. But when a guardrail is only 2 feet tall, according to Mr. Branch, the law asks whether a higher one was a reasonable alternative.

That is the classic design-defect test, and it is spelled out in the AI LEAD Act, introduced in September 2025 by Sens. Dick Durbin, Illinois Democrat, and Josh Hawley, Missouri Republican. The bill would treat AI systems as products and put primary liability on developers. A deployer would be treated as a developer only if it substantially modifies the system or intentionally misuses it. Courts would have to dismiss a deployer if a solvent developer is already in the case and the deployer isn’t otherwise liable. The bill has not moved beyond the Senate Judiciary Committee.

Courts are testing the chain, too. In May 2025, U.S. Senior District Judge Anne Conway let a Florida mother’s wrongful death suit against Character.AI proceed, along with claims against Google over its alleged role in developing the technology. Google said at the time that it “did not create, design, or manage Character AI’s app or any component part of it.” In January, Character.AI and Google agreed in principle to settle that case and others, with no admission of liability. OpenAI faces similar suits and has denied ChatGPT caused the death of 16-year-old Adam Raine.

States are drawing the line differently. Colorado’s 2024 AI law split duties between developers and deployers. After industry opposition and a lawsuit by Elon Musk’s xAI that the Justice Department joined, Gov. Jared Polis on May 14 signed a narrower replacement, effective Jan. 1, 2027. Under the new law, developers must give deployers documentation on known limitations and reasonably foreseeable harmful uses. Deployers must notify consumers.

President Trump’s December 2025 executive order ordered the creation of a Justice Department task force to challenge state AI laws, citing Colorado’s as an example.

The two guests landed far apart. “Let’s regulate specific harmful use cases,” Mr. Castro said. On the other hand, Mr. Branch said, “AI companies need to be held to the same standards that other industries are held to.”

New episodes of “The Dialogue” drop every Monday at noon on The Washington Times and YouTube. 

Watch previous episodes:

• Nancy Grace Reacts To The Lindsay Clancy Mistrial
• Should Religious People Watch Horror Movies?
• Are Americans’ Fears About Data Centers Overblown?
• Should Lindsay Clancy Be Held Responsible In The Killings Of Her Children?
• Should We Keep Funding Israel? America’s $3.8 Billion Question