Who Pays When AI Gets It Wrong: The Liability Question Wall Street Is Ignoring

The Question Nobody Is Asking About AI

Everybody is talking about how powerful AI is getting. NVIDIA just announced another $150 billion stock buyback, the largest in history, and they are rolling out new platforms to help developers monitor and govern AI behavior. The hype machine is running at full speed.

But I want to cut through all of it and ask the one question that actually matters: who is liable when AI goes wrong?

Because I promise you, something is going to go wrong.

The Jurassic Park Problem

NVIDIA’s new tool is called the Open Agent Safety Platform. It is designed to keep AI agents operating inside secure, governed environments. Developers can monitor behavior and make sure these agents follow rules.

Sounds great. But every time I hear this I think about Jurassic Park. They thought electric fences were going to keep the dinosaurs contained too. How did that work out?

I genuinely hope the engineers are right this time. But the history of technology tells us that guardrails get tested, and eventually, something gets through.

The Autonomous Driving Analogy That Changes Everything

Wall Street Journal columnist Andy Kessler made a point that I think is one of the most important frameworks for understanding where AI is actually headed. He compared OpenAI and Anthropic to Mercedes and BMW. Not because they are expensive or carry an attitude, but because they share a growing fear of liability.

Here is how the autonomous driving comparison breaks this down:

• Level Two autonomy means the car controls speed and steering, but you must keep your hands on the wheel and watch the road. If something goes wrong, it is legally your fault.
• Level Three autonomy means you can read a book or watch a movie. The car is actually driving. If something goes wrong, the manufacturer is on the hook.
• Level Four is what Waymo and robo-taxis are running, but they are geographically limited and weather dependent for exactly this reason.

Both BMW and Mercedes received approval in Germany in 2024 for Level Three. You know what they did? They said no thank you. They walked away from it. And the reason is simple. The moment a Level Three vehicle gets into a serious accident, the legal exposure is catastrophic. The media coverage would be wall to wall. The lawsuits would be massive.

So instead, they keep selling Level Two, where the driver is still technically responsible. Think about that. These companies have the technology to give you a better product and they are deliberately not deploying it because of liability exposure.

What This Means for AI Across Every Industry

This same logic applies to every AI application being built right now, including the ones touching your finances, your healthcare, and your legal decisions.

AI companies are currently covered by common law fiduciary responsibility. That is the framework that governs them. And the smarter operators in this space are starting to understand that the liability question is going to shape every major product decision going forward.

Anthropics CEO has said the industry must pace the frontier carefully. My translation of that is simple: they are watching the legal exposure very closely before they make their next move.

Here is what you need to take away from all of this:

• The most important filter for evaluating any AI product or service is asking who is responsible if it fails
• Companies building AI tools that touch your money or your safety are making deliberate choices about how much autonomy to give those tools, and those choices are driven by legal exposure, not consumer benefit
• The gap between what AI can do and what companies will actually deploy is going to be shaped almost entirely by liability law
• Investors piling into AI stocks need to understand that regulatory and legal risk is a major variable that is currently underpriced

The Bottom Line

NVIDIA’s $150 billion buyback is a headline. The real story underneath all of this AI excitement is a coming legal reckoning. When the first major AI failure hits, whether it is a financial recommendation gone wrong, an autonomous vehicle accident, or an AI agent that escapes its container and causes real harm, the liability question is going to define the entire industry.

I have been saying this for a while. Andy Kessler is now saying it in the Wall Street Journal. Start filtering everything you hear about AI through that one simple question: who is liable when this goes wrong? Your answer will tell you a lot about whether a given product, company, or investment is built on solid ground or on hype.