白宫Gold Eagle计划引发AI访问控制争议

White House program Gold Eagle is drawing attentio…

精选理由

白宫新规直接卡OpenAI和Anthropi的模型发布,跟国内Kimi K3的逆袭放一起看,特有意思。

AI 摘要

白宫Gold Eagle计划据报道赋予特朗普政府直接决定谁可访问OpenAI和Anthropic前沿模型的权力,此前由实验室自主控制。CNBC称上月政府以国家安全为由阻止了Claude Mythos 5和Fable 5的发布,谈判数周后才恢复访问。尽管白宫官员称测试自愿且发布决策权在公司,但实际操作显示更严约束。与此同时,中国Moonshot AI的Kimi K3在性能上追平Fable和GPT-5.6,并在至少一项独立基准上超越两者,David Sacks警告这可能导致美国输掉AI竞赛。

原文 · @koltregaskes

White House program Gold Eagle is drawing attentio…

White House program Gold Eagle is drawing attention for reportedly giving the administration more direct influence over who gets access to new frontier AI models from labs like OpenAI and Anthropic.

CNBC reports the Trump administration is now dictating which companies and agencies can access the latest frontier models, a role previously controlled by the labs themselves. Anthropic and OpenAI both ran their own partner programmes for cybersecurity models through Project Glasswing and Daybreak. Last month the administration blocked Claude Mythos 5 and Fable 5 over national security concerns, reinstating access only after weeks of negotiation. A White House official told CNBC that all testing is voluntary and release decisions rest with the companies, but the practical shift is clear.

The security angle is understandable given how quickly AI can accelerate both attacks and defences. At the same time, adding layers of approval on top of what labs already do voluntarily risks slowing US progress at exactly the moment competitors elsewhere face fewer self-imposed or external constraints. Chinese startup Moonshot AI unveiled Kimi K3 on Friday, which caught up to Fable and GPT-5.6 performance and even beat both on at least one independent benchmark. David Sacks called it concerning and warned this is how you lose the AI race if you bog yourself down while the rest of the world keeps moving.

How the clearinghouse actually operates over the next few months is the part worth tracking. The June executive order frames participation as voluntary, but last month's blocks and gating requests suggest the practical constraint may be tighter than the official line. https://t.co/W8VDm1VYnk