Trump Team Pushes US AI Abroad, Shelves Open-Source Curbs
TL;DR
- The White House is finalizing a voluntary frontier-AI framework asking labs to submit advanced models for government review 30 days before public release.
- The administration lifted export limits on Anthropic's frontier models while capping access to OpenAI's GPT 5.6 to a short list of US firms it approves.
- Roughly 80 percent of global open-source AI developers now build on Chinese models like Alibaba's Qwen, per Andreessen Horowitz research.
Washington seems to have picked a lane on open-source AI, at least this week. The New York Times reports that after officials weighed a more interventionist stance, including a Commerce Department draft aimed at Chinese open-source models that was ultimately halted inside government, the Trump administration has landed on promoting US-built frontier and open-weight models abroad rather than moving to constrain them at home.
The pivot arrives the same week the White House is finalizing its voluntary frontier-AI framework, tied to an early-June executive order and an August 1 deadline. Under it, AI companies would submit advanced models to the government for safety review 30 days before the model's public release, as CBS News reported. At the same time the administration has been openly asymmetric in how it treats individual labs, lifting export restrictions it had imposed on Anthropic's frontier models so they can be sold overseas, while reaching an agreement with OpenAI to limit access to GPT 5.6 to a handful of US companies and organizations approved by the White House.
The stakes are largely competitive. Research from Andreessen Horowitz cited by industry sources puts roughly 80 percent of the world's open-source AI developers on Chinese models such as Alibaba's Qwen, with US companies including Airbnb saying they are 'relying a lot' on Qwen because it is cheap and fast. If the strategy is to reverse that, the promotion push, which the AI Action Plan frames as 'export American AI technologies through full-stack deployment packages' led by Commerce, may matter more than any specific US safety rule.
The honest caveat is that this is a Washington tick-tock, and a lot of substance still turns on details the reporting does not pin down: how the finalized framework defines 'covered frontier models,' whether open-weight releases get the same 30-day review as closed ones, and who at Commerce actually owns the export push. The reporting also doesn't spell out the capability ceiling being discussed, the enforcement teeth if a lab ships without submitting, or what foreign governments are being asked for in return.
Read as a signal, the interesting move is the reframing. The administration has decided the answer to Chinese open source is not to slow US open source, it is to ship more of it, faster, with the federal government riding shotgun on the sales trip. Enterprises that were quietly prototyping on Qwen or Kimi now have a plausible reason to wait a beat and see what the US bundle looks like.
Originally reported by nytimes.com
Read the original article →Original headline: NYT: White House Pivots From Regulating Open-Source AI to Aggressively Promoting US Models Abroad