Model Distillation Undercuts Major AI Firms' Profitability

Business Insider reported July 8, 2026, that model distillation is eroding the profit logic for frontier AI labs by letting smaller models learn from outputs generated by systems from OpenAI, Anthropic, Google, and others. Bloomberg previously reported that major labs were sharing indicators through the Frontier Model Forum to detect adversarial distillation, while CNBC reported Alibaba restricted internal use of Claude Code after related accusations. For practitioners, the operational lesson is to evaluate distilled models independently: they may be cheaper and capable, but they can inherit hidden weaknesses, unclear training provenance, and weaker safety guarantees.