Nathan Lambert: open models may face US limits within months

Nathan Lambert's latest Interconnects post carries a blunt title and a specific timeline: open-weight models could face US regulatory restrictions within roughly six months. His reasoning is about triggers. Once open models cross a capability threshold he places in the range of GPT-5.5, Claude Opus 4.8, or GLM-5.2, he expects fear of frontier-level Chinese open models to push Washington toward controls, and several strong Chinese releases have already put that threshold in reach.

The sharper part of the argument is who he thinks is shaping the narrative. Lambert reads the distillation and security warnings coming from Anthropic as regulatory capture, arguing they rest on little public technical evidence. He points out that even Anthropic's private Mythos beta suffered unauthorized-access breaches, which undercuts the claim that keeping a model behind an API is inherently safer than releasing the weights. If a closed model can leak from inside a top lab, the safety case for banning open weights gets weaker.

His prescription follows from that. Rather than let the debate settle into a China-only framing, he wants Western firms like Meta and Microsoft to ship competitive open models now, so the policy question becomes about the whole ecosystem instead of a single country. It is a strategy piece as much as a policy one, from someone who tracks the open and closed model landscape closely.

Why it matters

If you build on open weights, this is a warning to plan for a possible regulatory shift on a short horizon, not a distant one. Whether or not the six-month figure holds, the argument that safety claims are doing lobbying work is worth weighing before you accept the case for restricting open releases.