In a live, meandering CNBC interview the CEO of data mining firm Palantir accused US frontier AI models of having “an effing insane “ business model which sees critical sector companies paying for tokens while potentially giving away intellectual property that gives them their competitive edge.

Initially invited to discuss Palantir’s new AI engine for the US govt built on Nvidia's AI infrastructure,Alex Karp’s meandering, almost 20-minute interview veered so far off script that it left even the presenters struggling to keep up with him.

“I feel like I’m going to get kicked out of the room,” he said at one stage.

Karp’s rant left viewers baffled: “I need a LLM to explain to me what the hell he just said,” one posted.

However, beneath the stream-of-consciousness delivery was a consistent argument – that AI users, particularly public and private businesses, should question what happens to their data when they use commercial AI models rather than just “chillaxing with their tokens” that “add little value.”

Karp: Frontier AI only benefits frontier AI

“Every single enterprise I deal with is livid,” Karp said. “They are saying ‘I am paying for tokens that create no value. Those people are stealing the weights and alpha [competitive edge] of my business.”

He claims that the “voice of American business” was being channeled through him, complaining that these complaints get “outsourced” to him, claiming companies prefer if they come from the “neurodivergent crazy person that apparently is on drugs, the one thing I don’t do.”

In recent months, some companies such as Uber have capped spending on AI tokens, while others moving away from premium AI models as rising usage costs make them harder to justify.

Karp adds that businesses who are paying millions in tokens are not given transparency over how their data is handled and he questions who ultimately benefits from the knowledge that enterprises contribute.

“Who owns the data? Where is it cached? Are the prompts secure?” he asked.

Karp also took several pot shots at OpenAI and Anthropic for misleading corporate partners and the public by trying to “oversell” the risks of AI while at the same time offering powerful models to companies and governments across the world.

“I’m not throwing Sam and Dario in the shade,” he later added, “but something has gone completely wrong,” he said.

Referring to Anthropic he says: “The ‘I’m going to trust you, you should trust me because I’ve never lied’ b.s. That just doesn't cut it at this level.”

What is Ontology?

Karp’s rant also appears to be a pitch for Palantir's own approach: rather than developing LLMs, the company has built what it calls “Ontology, an agnostic software layer that sits between a company’s data and an AI model, controlling what information the model can access and what actions it can perform.

Karp claims this allows customers to use any AI model they want without exposing sensitive information. He added that the technology is already being used by defence customers.

The conversation on this topic is live. Join in the discussion.

“Everyone who runs LLMs on the battlefield runs on top of our Ontology,” he boasted, before arguing the same trust concerns are now emerging across private industry.

However, the same trust issues LLM face have also been leveled at Palatir – especially outside its home territory.

Collaboration with the Israeli government and the Immigration and Customs Enforcement (ICE) agency, both of which are accused of human rights violations, has led to reputational damage.

Most recently, fear of data and national security leaks has prompted countries such as Spain to quietly ban Palantir's technologies from its systems.