Lambert warns open models could face US ban in 6 months

A blunt prediction from Nathan Lambert this week that is worth taking seriously even if you think the timeline is aggressive. Writing on Interconnects, Lambert says the most likely incoming policy action is to "ban or indefinitely delay any open-weights model meaningfully above the capability level in the range of GPT 5.5, Claude Opus 4.8, or GLM-5.2," and that on his read of White House discussions "this should be within the next 6 months."

His framing is not the usual safety debate. He argues the policy push is largely regulatory capture by closed model companies, and Anthropic in particular, who would gain economic ground if Chinese open-weight competitors got shut out of the US. He points to a mix of "blog posts and letters to representatives detailing what the Chinese companies are doing" as the mechanics of that campaign, and describes distillation, the other live policy fight, as one where "the only solutions on the table massively benefit the organizations pushing for it."

Two pieces of texture make the essay worth reading beyond the headline. First, he notes that "Chinese open-source models such as DeepSeek have a substantial lead over other available open models," which is why an uncoordinated US ban mostly hurts US researchers and startups rather than the actors it targets. Second, he points to Discord users gaining unauthorized access to Anthropic's own Mythos model to argue that "APIs continue to be jailbroken and accessed in unintended ways," so a policy premised on containing weights is fighting the wrong battle.

The honest caveat is that this is one researcher's read of private conversations, not a leaked draft executive order. The specific capability threshold, the legal mechanism, and the treatment of already-released weights are all things the piece flags as likely rather than confirmed. Take the six-month clock as Lambert's forecast, not a settled deadline.

His forward-looking piece is a proposed off-ramp: a US company shipping a competitive open-weight model would change the political calculus. He names Microsoft and Meta, whose businesses arguably benefit from commoditizing model weights, and says they "should release open-weight models ASAP." If either does, the story shifts. If they don't, the base case is that the frontier of open weights keeps moving offshore.