What happened
Anthropic, OpenAI, and Google report competitors are "distilling" their large language models, harvesting outputs at scale to train rival AI offerings. This process, using one AI model's outputs to improve another, mirrors the AI giants' own prior practice of scraping web content for model development. Anthropic CEO Dario Amodei states rivals turn billions of dollars of research into shortcuts, prompting Anthropic to tighten model access, though workarounds persist.
Why it matters
The cost of frontier AI model development faces direct pressure as distillation offers a cheaper path to comparable capabilities. For founders and research teams, this mechanism reduces the barrier to entry for advanced AI, potentially shifting competitive dynamics by devaluing proprietary training data and compute investments. This follows recent reports that LLM coding costs can exceed subscriptions tenfold, making cost-reduction strategies critical.
