The US Commerce Department has lifted the export controls it placed on Anthropic’s most advanced AI models, Claude Fable 5 and Mythos 5, ending an 18-day standoff between the company and the Trump administration.
Anthropic said the controls were removed on June 30, and that Fable 5 would return to users worldwide on July 1 across its Claude platform, website, Claude Code and Claude Cowork. Mythos 5, the more capable cybersecurity model that shares the same underlying system but ships with fewer safeguards, had already been restored on June 26 for roughly 100 vetted US organizations. Commerce Secretary Howard Lutnick said his department spent two weeks working with Anthropic to analyze and approve Fable 5.
The episode began when Anthropic launched the two models on June 9. Three days later, the government applied export controls requiring it to block access for any foreign national, inside or outside the United States, including Anthropic’s own non-citizen employees. Because the order took effect immediately and the company said it could not verify nationality in real time, it shut both models down for everyone.
The trigger, according to Anthropic, was a report from Amazon researchers describing a way to bypass one of Fable 5’s safeguards, prompting it to identify software vulnerabilities and, in one case, produce code showing how a flaw could be exploited.
Anthropic used its account to argue the flaw was less severe than the response suggested. The company said its testing showed that many weaker models, including its own Claude Opus 4.8, OpenAI’s GPT-5.5 and Moonshot’s Kimi K2.7, could identify the same vulnerabilities, and that the technique exposed only routine defensive security work rather than any unique offensive capability.
These are Anthropic’s characterizations of a dispute it was party to. Even so, the company said it retrained a safety classifier that now blocks the flagged technique in more than 99% of cases, routing blocked requests to Opus 4.8, at the cost of more false positives on legitimate coding tasks. Researchers at the Commerce Department’s Center for AI Standards and Innovation reviewed the updated safeguards, Anthropic said.
The Policy Push
Anthropic paired the resolution with two proposals aimed at shaping how such situations are handled in future. First, working with Amazon, Microsoft, Google and other partners in its Project Glasswing program, it is drafting an industry framework to score the severity of AI jailbreaks along four measures: how much capability the jailbreak unlocks, how broadly it applies, how easily it can be weaponized and how discoverable it is.
Second, it committed to deeper US government collaboration, including pre-release access to frontier models and their safeguards for designated agencies, faster sharing of threat intelligence, and dedicated compute for government testing. The company argued that any such government role should be written into durable, transparent regulation applied equally to all frontier developers.
The Unresolved Questions
The immediate crisis is over, but the governance questions it exposed are not. The controls were lifted in part amid criticism from tech executives and security researchers who warned that freezing a leading US model handed time to fast-rising Chinese open-weight rivals like Zhipu’s GLM-5.2.
Analysts noted that Washington reached for export controls, an improvised tool, rather than the voluntary review path set up under a June 2 executive order, which Fable 5 never went through. That leaves a central tension unresolved: pre-release government review may become standard for powerful models, which some will welcome as responsible practice and others will scrutinize for how much say Washington gets over which models ship, and whether an industry-drafted framework can double as public safeguard. For enterprises, the lesson many drew was the fragility of depending on a single model provider whose access can vanish overnight.
