Palantir Technologies CEO Alex Karp said frontier AI firms like Anthropic and OpenAI are attempting to "drug addict" their users [1].
This critique highlights a growing tension between the companies building foundational AI models and the enterprises that use them. At stake is the ownership of corporate intellectual property and the degree of autonomy businesses maintain over their own data and spending.
Karp said the business models employed by these AI labs are designed to create a dependency akin to addiction [3]. He said these firms are distilling the value of intellectual property from across the board while leaving the enterprises themselves with no control [4].
According to Karp, this process allows AI companies to extract and monetize the IP and data of their clients [3]. This dynamic creates a system where businesses may become reliant on a specific provider's ecosystem, making it difficult to migrate data or change providers without significant loss of utility.
"They want to drug addict us," Karp said [2].
He specifically targeted the way frontier labs handle enterprise data, suggesting that the current trajectory strips companies of their competitive advantages. By centralizing the intelligence derived from corporate data, Karp said these labs maintain a monopoly over the value created by the users' own information [4].
"Don't try to drug addict us," Karp said in a direct message to the leadership of OpenAI and Anthropic [3].
Palantir has positioned itself as an alternative to these models by emphasizing software that allows organizations to integrate AI while maintaining strict governance, and ownership of their internal data. Karp's warnings reflect a broader industry debate over whether AI should be a centralized utility or a decentralized tool controlled by the end-user.
“"They want to drug addict us."”
The conflict described by Karp represents a fundamental struggle over the 'AI stack.' While frontier labs provide the raw intelligence (the model), the data used to refine that intelligence often comes from corporate clients. If the model absorbs the client's unique intellectual property without giving the client control over that distilled knowledge, the AI provider becomes the sole owner of the resulting value, creating a high-barrier ecosystem that prevents corporate switching.
