If you tried using Claude’s newest models in the second half of June and got locked out with no real explanation, you weren’t imagining things. Anthropic’s two most advanced AI models, Claude Fable 5 and Claude Mythos 5, disappeared globally for almost three weeks after the US government stepped in. It’s one of the stranger stories to come out of the AI world this year, partly because of what triggered it and partly because of how little anyone said publicly while it was happening.
Here’s the timeline and why it actually matters beyond just being a weird headline.
What Happened, in Order:
Anthropic released Fable 5 and Mythos 5 on June 9. Both models share the same underlying system, but Fable 5 came with stronger built-in safety restrictions meant for general public use, while Mythos 5, which has fewer restrictions, was only handed to a small group of trusted partners for defensive cybersecurity work under a program called Project Glasswing.
Three days later, on June 12, the US Department of Commerce ordered Anthropic to cut off access to both models for any foreign national, anywhere, including Anthropic’s own foreign employees. Because the order took effect immediately and there was no fast way to actually verify someone’s nationality in real time, Anthropic made the call to suspend access for everyone globally rather than risk breaking the rule for a smaller group.
The trigger, it turns out, traced back to a report from Amazon researchers who found a way to get around Fable 5’s safety restrictions, essentially prompting it into identifying software vulnerabilities and, in one case, generating code that demonstrated how one of those vulnerabilities could actually be exploited.
What’s interesting is that when Anthropic tested this same technique against other models afterward, including its own older Claude versions and models from competitors, nearly all of them could reproduce the same behavior. So this wasn’t some unique danger specific to Fable 5; it was more of a borderline case where a safety filter that was designed to block a certain category of request out of caution ended up being bypassed.
On June 26, the government partially eased things, approving Mythos 5 access for a set of US organizations working on critical infrastructure defense. Then on June 30, the full export controls on both models were lifted entirely. Fable 5 came back globally the next day, July 1, across Claude’s platforms, including the website, the API, Claude Code, and Claude Cowork. Mythos 5 access has expanded to more approved US organizations, though it’s still not available everywhere; Australia, for example, still doesn’t have it back as of this writing.
Why This Actually Matters if You Use AI Tools:
It’s easy to read this as just another corporate drama story and move on, but there are a few things here that genuinely affect anyone using AI assistants day to day.
First, it shows how fragile access to frontier AI models can be. One government directive, and a tool that millions of people rely on for work can just disappear overnight with essentially no warning. If you’re building any kind of workflow or business process around a specific AI model, this is a reminder that continuity isn’t guaranteed the way it is with, say, a web hosting subscription.
Second, it highlights something a lot of people don’t think about, which is that AI safety restrictions aren’t just about stopping obviously harmful requests. Anthropic’s own explanation noted that some blocked behaviors are unlikely to actually be dangerous but get flagged anyway out of caution and that new, tighter safeguards introduced after this incident will likely catch a slightly higher number of harmless requests than before. In plain terms, if you’ve noticed Claude being a bit more cautious lately about certain technical or security-related questions, this is probably why.
Third, and maybe most interesting for anyone following the AI industry, this whole episode has raised real questions about precedent. Does every frontier AI release now need some kind of informal government review before it goes public? A former research director at an AI safety company was quoted describing this as a genuinely big deal precisely because of that question. It’s not just about Anthropic anymore; it’s about how the entire industry might operate going forward.
Where Things Stand Now:
Fable 5 is back and available to regular users through the usual Claude access points. Mythos 5 remains more restricted, limited mostly to vetted organizations working specifically in cybersecurity defense, and Anthropic has said it’s continuing to work with the government to expand that access over time. Worth noting too that the export controls letter reportedly left the door open for the government to reimpose restrictions later if it decides the situation calls for it, so this might not be the last chapter of this particular story.
For most everyday users, the practical takeaway is simple. If you used Claude and noticed weird downtime or missing model options in mid- to late June, that’s exactly what was going on. And if you’ve noticed the model being slightly more conservative on certain technical questions since, that’s a direct result of the tightened safeguards that came out of this whole situation.
