After nearly three weeks of restrictions, the US has lifted its export controls on Anthropic’s Mythos and Fable AI models, allowing Canadian users to access Fable again and partially restoring Mythos access to select US organizations.
- US lifted export restrictions on Anthropic’s Mythos and Fable AI models.
- Ban was initially imposed after Amazon researchers identified a jailbreak.
- Anthropic implemented new guardrails and safety classifiers before resuming access.
What happened
The US government imposed export restrictions on Anthropic’s Mythos and Fable AI models on June 12, following a report from Amazon researchers who demonstrated a method to bypass Fable 5’s security safeguards. This method, or ‘jailbreak,’ raised concerns about the exposure of software vulnerabilities. Although Fable 5 was designed as a more restricted version of Anthropic’s Mythos model, the jailbreak technique reportedly applied to several other AI models as well.
After conducting further testing, Anthropic confirmed no unique offensive cyber capabilities were discovered through this jailbreak in the Mythos or Fable models. Despite this, they introduced a new safety classifier intended to block attempts replicating the reported jailbreak method. Subsequently, the US government lifted the export ban, allowing Anthropic to resume service access for users globally, including Canadians accessing Fable and limited US access to Mythos.
Why it matters
The export restrictions had created significant uncertainty within the AI and technology communities, particularly in Canada where access to powerful models like Fable and Mythos is critical for cybersecurity research and innovation. The ban interrupted the availability of these tools, which were previously accessible under Anthropic’s Glasswing program that provides organizations with advanced AI capabilities.
Restoring access signals a renewed confidence from the US government in Anthropic’s security measures and a recognition that the reported vulnerabilities were not unique or inherently dangerous. For Canadian organizations, this means they can once again leverage Fable’s capabilities in cyber defense and related applications, supporting ongoing technological advancements and cross-border collaboration.
What to watch next
Anthropic is coordinating further with the US government to expand access to Mythos 5 for a broader set of domestic and international partners. This includes evaluating the safe rollout of models under the Glasswing program while maintaining tight controls against potential exploitation methods similar to the previously reported jailbreak.
Canadian authorities and stakeholders will be monitoring any updates on Mythos access availability and the practical impact of Anthropic’s new safety classifier. This will be particularly relevant for cybersecurity entities that rely on these models for vulnerability identification and defense mechanisms. Ongoing transparency and collaboration between governments and AI developers will remain key to managing risks associated with advanced AI export control.