Washington is moving fast on AI regulation. An executive order reviewing AI models is already signed. Congress has a draft proposal in circulation. The government may take equity stakes in frontier labs. Last Friday, foreign nationals lost access to Anthropic's most advanced models entirely. The authors, Nathan Lambert and Kevin Xu, argue the next target could be open source AI, and that banning it would be a serious, compounding error.
The case rests on three decades of evidence. Over 90% of the world's software runs on open source foundations. That base generated more than 8 trillion dollars in economic value before AI was a mainstream concern. Linux killed the Windows monopoly and now runs over 90% of cloud infrastructure. Android prevented iPhone from owning mobile outright. Today, Anthropic and OpenAI are consolidating closed, proprietary model power between them. Anthropic has already throttled its own model's capabilities when it detects use for competing model development. Open weight models are currently the only structural check on that concentration. The piece also addresses the security objection directly: transparency means more engineers can audit behavior and fix bugs, not fewer.
The argument that makes this worth reading in full is not the conclusion but the progression. Lambert and Xu trace open source from the 1983 free software movement at MIT, through specific corporate monopoly fights, to the current AI regulatory moment. The through-line is concrete and the precedents are named. Multiple mainstream outlets rejected this piece before publication. That context matters for understanding what the AI policy conversation is and is not currently willing to consider.
[READ ORIGINAL →]