Karp: AI companies may need nationalising to cap their liability

Key points

Key takeaways from CNBC interview with Palantir co-founder and CEO Alex Karp (September 2026):

Liability is the real debate. Karp argues the public fight over AI safety is not what it looks like — the underlying question is liability. Frontier labs' philosophical frame, that they are deserving of having all the IP in the world, would require severe liability protection, and only one institution can provide it: the US government.

Nationalisation is the endgame. Karp says the labs may have concluded that they have to be nationalised, because if they are not, "every single one of my clients is going to sue." Once an AI business presupposes unlimited risk against clients and the world, the only exit is migrating the business to the government.

Clients are livid about IP capture. Palantir's customers believe their "alpha" — how they run their business, their data — ends up in models that competitors can also access, and they paid for it. Karp describes the mildest client complaint as: "How can they tell me not to wear a condom?"

Token discounts are a data play. He argues tokens are deliberately underpriced so labs can access enterprise IP and make their models better, with no honest conversation or disclosure about it — the driver behind the sovereignty push Palantir helped launch with partner Jensen Huang of Nvidia.

Open-weight momentum in classified work. Post-training with open-weight models in the classified/defence world is taking off across America, Karp says, because it is cheaper, ends up safer, and lets customers control their own future rather than outsourcing it to a closed model.

On Dario Amodei. Karp praises the Anthropic CEO as a shrewd, ethical operator who came "from fifth behind to number one," but notes safety talk tends to surface when the company is doing less well.

Read more: CNBC Television on YouTube · CNBC.com