Palantir CEO Alex Karp has criticised frontier AI companies including OpenAI and Anthropic, accusing them of encouraging businesses to become increasingly dependent on their AI systems.
"We have people trying to drug addict us to a future they believe they control," Karp said in a CNBC interview, referring to frontier AI companies such as Anthropic and OpenAI.
Karp argued that businesses were being encouraged to spend heavily on AI models and computing without necessarily retaining control over their data, intellectual property or the value generated by those systems.
He said companies were being pushed towards a future where their businesses were not profitable, jobs were lost and their adversaries gained an advantage. Karp has previously criticised the business models of OpenAI and Anthropic, particularly their token-based pricing models.
"Tokenmaxxing" refers to using large amounts of AI computing, often through extensive prompting or autonomous AI agents, without clear evidence that the additional spending is generating meaningful business value.
Karp said enterprises were questioning why they should pay for AI services while giving up control over systems needed to develop their businesses and retain their value.
OpenAI and Anthropic have said customer data is secure and is not used to train their AI models. OpenAI has also said businesses must explicitly opt in for their data to be used for model improvement.
Karp has backed open-weight AI models, which allow developers and businesses to access the models' underlying weights, customise them and run them independently rather than relying entirely on a provider's service.
Last month, Palantir joined Nvidia, Microsoft and other companies in signing an open letter arguing that open-weight AI models are important for national security.
The debate comes as the performance gap between open-weight and closed frontier AI models narrows.
A recent study by the UK's AI Security Institute found that Chinese open-weight models such as Z.ai's GLM-5.2 and DeepSeek V4-Pro were performing at levels comparable to US frontier closed models released four to seven months earlier.
During most of 2025, the estimated gap was six to 10 months.