Artificial intelligence "distillation," the practice of training a smaller model to copy the behavior of a larger one and inherit its capabilities at far lower cost, is at the center of a sharply worded argument. Alex Karp, chief executive of Palantir, renewed his attacks on America's frontier AI laboratories, calling them companies "trying to drug addict us" while also defending Chinese AI developers from criticism over the same practice.
What Karp is saying about frontier labs
Karp's phrase "drug addict" implies design intent, not accident. His argument is that the leading AI laboratories are engineering dependency, pulling users into systems built to hold them. He did not name specific companies.
The term "frontier lab" refers to organizations competing to build the most capable AI systems available. These are expensive, large-scale operations sitting at the outer edge of what the technology can currently do. Karp's criticism is that their position rests on extraction, specifically on absorbing intellectual property from across the world to build the models that now power their products.
His defense of Chinese AI distillation
Critics have argued that Chinese AI developers distill American frontier models: they take the outputs of sophisticated, expensive American systems and use those outputs to train cheaper alternatives that inherit much of the performance. The practice allows a developer to shortcut years of training investment.
Karp pushed back on blaming China for this. The frontier labs, he said, "distilled all the value of IP, everywhere." His point is structural. If American AI companies built their power by harvesting intellectual property broadly, they are poorly positioned to object when that technique is applied to their own models.
The renewed attacks continue a pattern Karp has kept up over time, placing Palantir in deliberate contrast to the industry he criticizes.