"We will have done something wrong if we just accept this"
Global and domestic responses to the ongoing Fable dispute
Dispatches from Mitch
White House AI policy is global AI policy
The Fable situation has the world more nervous than ever that our future might be decided by just a handful of people who are almost all Americans: the CEOs of the leading AI companies, and the President of the United States.
Set aside, for now, the fact that an unchecked race to artificial superintelligence eventually kills everyone, no matter who gets there first: There has long been concern that American AI companies could build up an insurmountable lead over their overseas rivals, and that this would eventually enable the U.S. to dictate global policy.
The international reactions to the Fable shutdown are, therefore, lingering over the fact that the White House’s order to pull the plug on Fable/Mythos was executed as an export control directive. Whatever the actual intent, the stated intent was to deny all non-Americans access to the world’s strongest AIs. For now this means cutting off access to all Americans, too, but will that always be the case? Few are willing to bet on that.
The result is one of the worst outcomes: Because the U.S. hasn’t coordinated a global pause to the AI race, countries that were content with watching from the sidelines now feel compelled to join it, lest they become dependent on an unreliable partner with a track record of cutting off their access.
Canadian Prime Minister Mark Carney took a diplomatic tone when commenting on the Fable shutdown, but you can hear the frustration in remarks reported by the Washington Post:
The situation we’re in collectively right now with Mythos and Fable is something that can happen with overreliance on certain models. [...] Nobody has done anything wrong in the situation. But we will have done something wrong if we just accept this, don’t take the lesson, don’t build out and diversify. [...] It is never a good idea to have one option.
You know you’re doing American AI policy wrong when you drive Canada to compete with you.
With France, I would expect this as par for the course, and President Macron hasn’t disappointed. He spent 45 minutes talking to Carney about AI on Friday night. Separately, he criticized the U.S. for using AI as a “power tool” and talked up the potential for an AI alliance with India. And French presidential candidate Gabriel Attal tweeted that “The AI war has already begun.”
Meanwhile, in China, shares in domestic AI companies are surging in anticipation of growing demand from customers determined to diversify their AI providers. Per CNBC, developer Zhipu is up 33% since the Fable ban. The company was quick to recognize the opportunity; in a statement, they wrote:
Cutting-edge intelligence should not belong to only a few, nor should it be withdrawn at any time. It should be open, available, extensible and built to serve every developer.
You know you’re doing American AI policy wrong when you’re pumping up the Chinese AI industry.
As Zhipu hinted at, the strongest hedge against governments cutting you off from AI is to switch to models you can host yourself. As Aaron Levie, the CEO of Box, points out:
The big winner in all of this is going to be open weights models. This is a huge win for the field, as a risk that was entirely theoretical and untested 2 days ago (that a model could be pulled back), now has a new precedent that’s been set.
You know you’re doing American AI policy wrong when you’re making people see open-weights as a must-have feature, shrinking your future ability to take dangerous models offline.
The thing is, I’m with researcher Jeffrey Ladish here that it’s great that the White House has shown how quickly and decisively it can respond to perceived AI threats. As he put it:
If they’re just targeting Anthropic out of spite, that is bad. But if they’re genuinely worried about the national security threat posed by the model, then I think their actions are in the right direction given their understanding. I want them to be more transparent about their reasoning, but I think it’s possible they’re just freaked out.
If the White House had just told everyone, “Hey, we’re concerned about this model and are shutting it down,” instead of saying, “Only Americans are allowed to access this model we’re concerned about bad actors having access to,” Fable would still be offline to everyone today, but I would be writing a very different dispatch.
America’s dominant position in AI right now means that White House AI policy is global AI policy, whether it means to be or not. And right now, that policy is promoting proliferation that will make everything harder later.
“Government messaging 101”
On the home front, finger pointing and finger wagging continue over the White House’s choice to force Anthropic to shut off all access to its pair of frontier AI models, Fable and Mythos.
Fox Business’s Edward Lawrence relayed the perspective of a “senior administration official” who claims Anthropic had been “reckless” and had brushed off the government’s initial concerns.
Lawrence’s article dutifully reports the counterclaims and counterevidence to these assertions, which I also covered in yesterday’s dispatch. Lawrence seems to have been given the impression that the executive branch wanted to make an example out of the AI company. He writes:
A Washington, D.C., tech policy expert tells FOX Business this is “government messaging 101” – that if a company is leading the conversation on AI safety, then appears reluctant to address any safety concerns, the administration will take a tough stance.
It’s hard not to see this as evidence that Anthropic’s “mistake” was having voiced concerns about safety and treating its new models as needing strong guardrails.
I think it would be better if Anthropic hadn’t made dangerously powerful new models in the first place, but you shouldn’t shoot messengers if you want to keep getting messages. And I dearly hope the White House wants to keep getting messages.
Coverage from the Washington Post this morning shared perspectives from DC insiders like former White House AI adviser Dean Ball, who tweeted:
Make no mistake: post-Mythos, the United States has a licensing regime for AI. It’s just informal, with no consistent rules or firm boundaries on state power or public transparency.
Unnamed sources familiar with the discussions between the government and Anthropic were also cited; in their view, the White House’s recent executive order that requested voluntary access to new models and swore off any kind of AI licensing regime now looks “performative.”
The Wall Street Journal’s Amrith Ramkumar and Robert McMillan (continuing their streak of strong reporting on this story) shared that some of Anthropic’s technical leads were to fly to DC to meet with government security experts today and hash things out.
The Journal adds that a group of “cybersecurity notables” are rooting for them. In an open letter, the researchers call for a lifting of the export controls keeping them from their strongest tools. From the letter:
This action has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.
To be clear, there’s always a “real risk” because today’s AI models, and the technology behind them, are so poorly understood. Nobody can say for sure what Fable and Mythos are capable of with the right prompts. The line between “tool” and “independent actor with uncertain agenda” gets fuzzier with every new release.
But yes, the growing weight of evidence suggests that threats from AI were not the White House’s main motivation Friday.
The analyses and opinions expressed on AI StopWatch reflect the views of the individual contributors and the sources they cover, and should not be taken as official positions of the Machine Intelligence Research Institute.