Hey there,

Last week we covered the release of Anthropic's new Mythos-class model, Fable 5. Later that same day, Anthropic suspended access to the model. This wasn’t planned but was instead a reaction to a directive by the US government. So what happened? Will we get Fable 5 back in the future? We’ll answer these questions and explain the situation this week. 

The directive that killed Fable 5

On the evening of June 12, just three days after the release of Fable 5, Anthropic was ordered to suspend access to both models for all foreign nationals. This referred to people both inside the US and abroad and even included Anthropic’s own employees. The government cited national security concerns as the cause of the export control directive. This type of specific user filtering wasn’t practical or possible for Anthropic, so they instead decided to revoke all access to the model for the time being. While not released publicly, Mythos 5 was also impacted by the directive. Access to older models like Claude Opus 4.8 continued uninterrupted. 

What triggered the directive?

Anthropic has explained that the directive was a result of the government believing that a method to jailbreak Fable 5 was found. The jailbreaking centered around bypassing the safeguards meant to limit Fable 5 from completing any cybersecurity tasks. The company says it has reviewed a demonstration of the technique used and confirmed it was able to identify several minor vulnerabilities that were previously known. However, Anthropic claims that these vulnerabilities could have been identified by other publicly available models. 

Where things currently stand

The government hasn’t released a detailed technical explanation aside from the initial export control directive. However, it's clear that they believe that the model is a national security concern that requires immediate action. 

Despite their disagreement, Anthropic is complying with the order. In their statement last week, Anthropic argued that the minor jailbreak should not be grounds for recalling an already public model. They go on to say, “If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”

In their public release, Anthropic repeatedly pushes back on the notion that Fable 5 is dangerous. They explain how they worked with both the US and UK governments as well as internal and third-party red teamers to test for any major, “universal” jailbreaks and how none were found after spending thousands of hours. 

The future of Fable 5

As of this writing, Fable 5 is still offline. Despite Anthropic's explanations and further assurances, the export control directive is still in effect, and it seems no mutual consensus has been reached. Behind the scenes, both sides are talking. Anthropic’s senior technical staff met with officials in Washington this week, and Anthropic personnel have expressed confidence that the model will be back soon. 

My Thoughts

The export control directive and suspension of Fable 5 really took me by surprise. I was just starting to integrate the model into my workflows when access was abruptly cut off. As someone who follows AI news closely, I definitely understand why the government is stepping in to address an issue that it feels is potentially dangerous. Making these powerful models public always has risks, so they need to be continuously assessed and monitored. That being said, it seems like the original jailbreaking incident was rather small and can apparently be replicated using other models already widely available, so the directive may not actually be having its intended impact. The long-term impacts of this directive will depend on how long it lasts: a short, one to two week period of additional testing and double-checking will be a minor roadblock, but an extended pause will leave a lasting mark on Anthropic and other frontier labs.