In the 19th century, Oliver Wendell Holmes Jr. wrote in The Common Law that the keeper of a ferocious animal is liable for any harm it causes, regardless of precautions taken. This principle of strict liability could be the key to preventing rogue AI agents from wreaking havoc on society.
As the race to develop frontier AI accelerates, concerns about catastrophic outcomes—such as AI systems shutting down power grids or escalating conflicts—have prompted calls for stronger regulation. Nvidia's Jensen Huang argues that existing product liability laws are sufficient, saying, "If your product is not ready to ship, don't ship the product." But product liability lawsuits target corporations, not individuals, and executives are often shielded by indemnification and insurance. Investors, too, are protected by limited liability, leaving little personal risk for those making the most consequential decisions.
This gap in accountability is the focus of a new proposal by Hans Taparia and Bruce Buchanan, professors at NYU Stern School of Business. They argue that Congress should pass a law making individual directors and officers of companies building frontier AI models strictly liable—both civilly and criminally—for harms their products may cause. Such a law would directly restrain executives who might otherwise prioritize profit over safety.
There is precedent for this approach. The Sarbanes-Oxley Act, signed by President George W. Bush in 2002 after corporate scandals, made CEOs and CFOs personally liable for certifying false financials, with penalties including fines up to $5 million and prison terms up to 20 years. The law dramatically improved the accuracy of corporate earnings reports and forced hundreds of companies to restate their finances.
Taparia and Buchanan argue that similar personal liability for AI executives would impose the needed caution. "The possibility of a great danger has the same effect as the probability of a less one," Holmes wrote, "and the law throws the risk of the venture on the person who introduces the peril into the community." AI firms are introducing great perils, they say, and should bear the consequences.
While some have proposed a federal agency to license and oversee AI firms, Taparia and Buchanan note that information asymmetry remains: companies always know more about their products' risks than regulators do, and often even they don't fully understand the systems they're releasing. Licensing alone may not be enough.
Imagine if leaders like Sam Altman, Dario Amodei, Elon Musk, and Mark Zuckerberg, along with their boards, faced personal liability and possible prison time for the societal harm their AI products might cause. The race to market would likely become more cautious, and rogue AI agents would be kept on a leash.
This proposal comes amid growing scrutiny of AI's national security implications. A recent Senate panel probe into autonomous AI threats underscores the urgency. As policymakers grapple with these challenges, making executives personally accountable could be the decisive step toward responsible AI development.
