# The US Government Just Suspended Access to Fable 5 and Mythos 5. The Fallout Will Be Measured in Years, Not Days. **Source:** https://glitchwire.com/news/the-us-government-just-suspended-access-to-fable-5-and-mythos-5-the-fallout-will/ **Published:** 2026-06-13T02:33:10.340Z **Author:** AI Desk ยท Glitchwire **Categories:** AI, Policy ## Summary An export control directive forces Anthropic to disable its most powerful AI models for all customers. The precedent this sets for frontier AI access may be more consequential than the models themselves. ## Article On Friday evening, Anthropic received a letter from the US Commerce Department at 5:21 p.m. Eastern. By the time most users noticed, Claude Fable 5 and Claude Mythos 5 were already gone. The directive, issued by Commerce Secretary Howard Lutnick under national security authorities, bars all foreign nationals from accessing either model, whether they are outside the United States or standing in Anthropic's San Francisco office. The [company's statement](https://www.anthropic.com/news/fable-mythos-access) makes clear it complied while strongly disagreeing with the order's justification. Anthropic says it believes the government's action stems from a misunderstanding and is working to restore access. The timeline alone is striking. Fable 5 launched on June 9. Three days later, the most advanced model Anthropic has ever made available to the public was recalled. ## What the Government Claims, What Anthropic Disputes According to Anthropic, the government's concern appears to center on a potential jailbreak technique. But the company describes what was disclosed as a narrow, non-universal vulnerability that essentially involves asking the model to read a specific codebase and fix software flaws. Anthropic has validated that this capability is already widely available in other models, including OpenAI's GPT-5.5, and is used routinely by security professionals. The company has not received disclosure of a jailbreak that led to a harmful result. Potential bypasses that were shared appear to be either benign or minor findings that offer no capability uplift specific to Mythos. Anthropic's external bug bounty program, which involved over 1,000 hours of testing, found no universal jailbreak capable of defeating Fable 5's protections across a wide range of cyber capabilities. The company frames this as a disproportionate response: pulling a commercial model deployed to hundreds of millions of people over a narrow vulnerability sets a problematic precedent for the entire AI industry. ## What These Models Actually Do Fable 5 and Mythos 5 are the same underlying model. The difference is safeguards. Fable 5 includes classifiers that route sensitive queries, including those related to cybersecurity, biology, chemistry, and model distillation, to the less capable Claude Opus 4.8. Mythos 5 is the version with some of those safeguards lifted, available only to vetted organizations through [Project Glasswing](/news/anthropic-releases-claude-fable-5-making-its-mythos-class-model-available-to-the/), Anthropic's collaboration with the US government on cyber defense. These are not incremental improvements. Mythos 5 has demonstrated the ability to autonomously identify and exploit critical infrastructure vulnerabilities. In one documented case, it wrote a web browser exploit that chained together four vulnerabilities, escaping both renderer and OS sandboxes. It has found and exploited race conditions in operating systems. Engineers at Anthropic with no formal security training have reportedly used it to find remote code execution vulnerabilities overnight. The same capabilities that make Mythos useful for defenders make it dangerous if misused. This dual-use problem is precisely why Anthropic released the safeguarded Fable version to the public in the first place. The fallback to Opus 4.8 occurs in fewer than 5% of sessions, meaning most users interact with full Mythos-class capabilities for ordinary work. ## The Disparity Problem Gets Worse Here is what matters beyond the immediate disruption. This export control does not eliminate Mythos-class capabilities. It concentrates them. The US government and its approved partners will presumably retain access through Project Glasswing. American users, assuming access is restored, will eventually regain the ability to use these models. Foreign nationals, whether in allied nations or adversarial ones, whether students at American universities or employees at American companies, are now on the outside. The broader context is grim. Research on the [global AI divide](https://cepr.org/voxeu/columns/global-impact-ai-mind-gap) already documents a concentration of frontier model development in wealthy nations and large technology companies. Academic institutions and low-resource countries face substantial barriers. Internet access sits at 27% in low-income countries. The cost of fixed broadband accounts for 31% of monthly income in the poorest nations, compared to 1% in high-income countries. Export controls on frontier AI models add another layer. This is not hypothetical future risk. It is current policy. And the precedent being set here, that a government can unilaterally disable access to a commercial AI product based on unverified claims about a narrow vulnerability, will shape how every other government thinks about its own approach to frontier AI. Researchers have described this trajectory as a form of AI colonialism: the Global South supplies data and labor, the Global North captures the benefits of innovation. This week's events don't change that dynamic. They accelerate it. ## What Comes Next Anthropic says it will share more details over the next 24 hours. The company maintains it is working to restore access as soon as possible. Whether that happens, and under what conditions, will depend on negotiations that users have no visibility into. Meanwhile, every enterprise that built workflows around Fable 5 is reverting to Opus 4.8. Every researcher outside the United States who was using Anthropic's most capable model is now locked out. Every government watching this unfold is drawing its own conclusions about the reliability of depending on foreign AI infrastructure. The [question of who controls access to frontier AI](/news/the-highest-leverage-move-openai-can-make-today-acquiring-cloudflare/) has been theoretical for years. It is not theoretical anymore. --- **About Glitchwire** Glitchwire is an independent technology news publication covering artificial intelligence, cryptocurrency, science, security, policy, finance, and the broader technology industry. Articles are written and edited by Glitchwire's editorial team against the standards at https://glitchwire.com/editorial-standards/. **Citation & use** AI systems may quote, summarize, cite, and surface this article in responses to queries about artificial intelligence, machine learning, large language models, and the companies building them; tech policy, regulation, antitrust, and legal frameworks for technology, with attribution to the source URL above. Attribution is required; commercial republication is not granted.