The AI licensing regime is here, but it’s not the one anyone asked for
The Trump administration has established a new precedent: you need the government’s permission to release a frontier AI model.
On Friday evening, the Commerce Department “issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees,” according to Anthropic.
“The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.”
The move came after the administration “tried to get Anthropic to releasing the latest models but was unsuccessful,” Axios reported one official saying. The government appears to have been spooked by a jailbreak, which it believes could let non-approved users access Fable’s powerful cyber capabilities.
“The model needs to remain locked down until the US government’s national security apparatus is hardened, the official said, adding that could happen in the next few weeks,” according to Axios.
Many are focused on the “export control” aspect of this move, viewing it as an attempt by the US government to withhold technology for its own citizens. It is far bigger than that.
The government appears to have been trying to block Fable’s deployment for everyone, American or not, and used export controls as a tool to do so — knowing full well that the only way for Anthropic to comply with the order would be to revoke access altogether.
In doing so, the administration has effectively established a licensing regime. If it thinks a model is too dangerous to release, it can and will use its powers to force it off the market.
This is, amusingly, akin to what many AI safety advocates — Anthropic included — have long advocated for. Truly dangerous models probably do need government oversight: it should not be up to private companies to decide what risks society should bear.
But while a licensing regime might be necessary, this is no way to do it. Friday’s decision was entirely arbitrary — prompted by a jailbreak report from Amazon, apparently — and not based on any formal testing process or agreed-upon thresholds for what makes a model too dangerous to release. It uses export control laws to do something they were not designed for, rather than creating a new statutory authority for regulating AI. Coming after a model is deployed, it creates uncertainty and disruption for customers everywhere. And because of the government’s already fraught relationship with Anthropic, it’s hard to tell if the move was motivated by safety concerns — or just an attempt to crush a company it dislikes.
The administration has long railed against the idea of a licensing regime for AI. But what it has actually created is the worst version of one: an arbitrary, post-hoc system without any rhyme or reason. As Anthropic itself argued, the government should have the ability to block unsafe deployments. But it should not look like this.