Palantir vs. Anthropic: Alex Karp Says the AI Safety Debate Is About Liability

Palantir CEO Alex Karp is challenging the AI industry's safety debate with a different question: who gets sued when AI goes wrong?

Karp argues that growing calls for government involvement in frontier AI may partly reflect the enormous legal risks facing companies building the world's most powerful models.

Speaking on CNBC's Squawk on the Street, Karp specifically targeted Anthropic, arguing that AI labs could eventually face lawsuits from companies claiming valuable proprietary information entered their models and potentially benefited competitors.

Karp said some Palantir customers are already worried about their corporate knowledge—or what he calls their business “alpha”—being absorbed by outside AI systems. He suggested that sufficiently large liability risks could eventually push frontier labs toward government protection or even nationalization.

Karp Turns AI Safety Into a Liability Fight

Karp argues that AI developers should remain exposed to civil and potentially criminal liability when irresponsible development causes harm.

“The first line of defense is you're liable for your own actions,” Karp told CNBC.

The corporate-data issue is already becoming tangible. Palantir has reportedly pushed Anthropic for stronger zero-data-retention guarantees, while Nvidia and Booz Allen Hamilton have also placed restrictions on the use of some frontier models for sensitive work. Anthropic and OpenAI say they do not train on enterprise customer data by default unless customers opt in.

Anthropic Sees a Different Risk

Anthropic CEO Dario Amodei has focused instead on the technology itself, warning that AI capabilities could advance faster than researchers can understand or control them.

His push to pace frontier development has already shaken the technology sector, with AI stocks selling off after his warning. OpenAI and Elon Musk have also supported parts of the slowdown argument, widening the debate over AI restraint.

Karp is effectively adding a third issue alongside AI safety and regulation: legal responsibility.

And as businesses put increasingly valuable data into AI systems, the question may become harder to avoid.

The next major AI fight may not simply be about how powerful models become, but who pays when they cause damage.