SITUATION ANALYSIS: Fable and Export Control
Last Friday, at 5:50 PM PT, Anthropic announced that the US government had issued an export control directive to suspend access to Claude Mythos 5 and Fable 5 to any foreign national, regardless of location, including employees of Anthropic. As a result, Anthropic was forced to immediately pull Mythos and Fable from all customers, just three days after their release, until citizenship verification can be implemented. Though frontier models have been withheld from the public for safety testing before, this is the first time that a frontier model has ever been de-deployed after its release. It’s also possible that foreign national AI researchers would be banned from working on frontier models, which would be a major disruption to the industry.
The government’s order came after they were warned by Amazon CEO Andy Jassy that Amazon was able to jailbreak Fable. The jailbreak in question was apparently just asking the model to find security vulnerabilities, something white-hat vulnerability researchers do all the time, and something that works on other companies’ models as well. It was not a universal jailbreak that would remove all safeguards from the model. Fable is also likely the most jailbreak-resistant model ever — in fact, its safeguards were so strong that they blocked access to the model entirely for almost any query related to bio or cyber. Commerce Secretary Howard Lutnick said the government feared the models being used for military intelligence in adversary countries.
On a micro level, it’s unlikely Fable will remain inaccessible to Americans for long. We’ve seen this pattern before: after market close, especially on a Friday, the Trump administration announces a major, unprecedented policy decision. X reacts in shock; many denounce it as arbitrary, capricious, foolish. Then, after a panicked intervention from some administration insider, perhaps warning the president of the policy’s impact on the stock market, the decision is rolled back or otherwise defanged, and the market opens up again. This has happened with tariffs, threats to fire Jerome Powell, and the designation of Anthropic as a supply chain risk in March. Anthropic has sent a delegation to the White House to negotiate, and prediction markets expect Fable access to be restored to customers within 1-2 weeks.
On a macro level, the influence of government over the AI industry continues to rapidly advance. In just a month we’ve gone from the Trump administration touting its laissez-faire attitude toward the technology, to implementing a voluntary (for now) frontier model pre-deployment check, to ordering export controls on a frontier model and forcing it to be broadly recalled. There’s no reason to expect this trend to stop, as models get more powerful and government gets more AGI-pilled.
And as for countries that aren’t the US or China, now is the time to start building sovereign AI. In 2021 you could build a frontier AGI lab with a few hundred million. In 2023/24 it took a few billion. That number is rapidly rising, and now only a few large nation-states are capable of really doing it — Japan, Germany, Britain, India, France, and a few others. Europe 2031, a scenario forecast released last week, predicted US technological dominance and European indifference to sovereign AI would make the continent economically irrelevant. In the scenario, the US restricts allied countries from accessing frontier models in 2029. In real life, the first warning shot was just fired three years ahead of schedule.
We will keep monitoring the situation. Via
@theojaffee
SITUATION UPDATE: Anthropic confirms it is disabling Mythos and Fable 5 for all customers to comply with a US government export control directive.
The company says it believes the order is a misunderstanding and is working to restore access.