Commerce conditioned the export lift on safety testing cooperation after Anthropic's chatbot was flagged as a cyber risk. The stronger safeguards that cleared the bar now sometimes block the routine coding tasks developers rely on.
Anthropic's consumer chatbot Fable 5 is once again available to users outside the United States, ending months of geographic restrictions the Commerce Department imposed over cyber-risk concerns. The model came back online globally after Anthropic agreed to cooperate with US government safety testing, but the safeguards that won export clearance now sometimes block the routine coding tasks developers rely on.
On June 12, 2026, the Commerce Department ordered Anthropic to cut off non-US access to Fable 5 and its more advanced sibling Mythos 5, citing concerns that the models could be weaponized against US infrastructure by China, Russia, or other countries of concern. Anthropic complied, shutting off all non-US access because the company said it had no way to block users by country.
Commerce framed the action around Mythos 5, which Anthropic's own communications had described as "uniquely attractive to malicious actors" for cyberattacks and "more effective than any other model" at finding and exploiting software vulnerabilities. That framing carried weight until independent researchers began testing comparable systems.
The case for treating Mythos as uniquely dangerous started to crack when Amazon researchers identified a bypass method that coaxed the model into surfacing software vulnerabilities it had been trained to refuse. Anthropic's own follow-up testing then found that less advanced models, including OpenAI's GPT-5.5 and Moonshot's Kimi K2.7, could identify the same weaknesses, according to Ars Technica. The result quietly softened the claim that Mythos-level offensive cyber capability was singular enough to justify a hard export cutoff, though Anthropic still describes Fable 5 as offering no such unique offensive capabilities in its current form.
To unlock the export hold, Anthropic committed to cooperation with Commerce on safety testing conditions, as Commerce Secretary Howard Lutnick signaled in a public post on X. The company rebuilt Fable 5's safeguards, which Anthropic says is built on the same underlying model as Mythos 5 but offers no unique offensive capabilities. According to Anthropic, the new protections are the strongest it has applied to any model and now block the bypass method in more than 99 percent of cases.
The same hardening carries a product cost. Anthropic has acknowledged that the stronger Fable 5 safeguards sometimes block routine coding and debugging prompts, a trade-off the company treats as a real consequence of the safety work rather than a footnote. Al Jazeera's reporting and Fox Business both note the same constraint: the model is now harder to weaponize but also harder to use for the everyday programming work that made it a daily tool.
The deal raises a question worth tracking as more frontier models hit similar regulatory pressure points: when a government conditions export clearance on safety testing, who pays for the hardening in practice? In this case the cost shows up not as a price tag but as a coding assistant that occasionally refuses legitimate work.