APREX · EST. 2026
A united group of veteran cybersecurity experts has sent a formal protest to the White House, demanding that export control restrictions on Anthropic's advanced AI models Fable 5 and Mythos 5 be lifted immediately. According to TechCrunch, the group argues that the ban undermines the ability to protect critical software and digital systems — and that it is defenders, not attackers, who are hit hardest.
What Happened?
The US government recently issued an export control directive requiring Anthropic to suspend foreign users' access to Fable 5 and Mythos 5 on national security grounds. Because Anthropic had no technical solution ready to geographically segment users on short notice, the company chose to shut down both models for all users globally — including Americans.
The immediate trigger is reported to have been researchers discovering a method to bypass Fable 5's built-in safety guardrails, potentially allowing the model to behave like the less restricted Mythos 5. Anthropic disputes the severity of the finding, however, arguing that the jailbreak in question is narrow, non-universal, and only produces vulnerabilities at a level already achievable with other publicly available AI models.
Using AI to fix code and generate test scripts is routine defensive work — not a circumvention of safety guardrails that should trigger export controls.
Mythos 5: Built for Defenders
Mythos 5 is not a standard consumer model. It shares the same technical foundation as Fable 5, but with certain safety guardrails lifted to make it usable in professional cybersecurity work. The model was originally rolled out through what is referred to as "Project Glasswing," a collaboration between Anthropic and the US government aimed at cyber defense workers and critical infrastructure.
Anthropic has itself described Mythos 5 as the model with the strongest cybersecurity capabilities of any model in the world — a characterization that underscores what the defense sector has now lost access to.
The Experts: Defenders Are Paying the Price
Luta Security CEO Katie Moussouris is among those who have reacted most strongly. She points out to TechCrunch that using AI to fix code and generate test scripts is an entirely ordinary and valuable defensive task — not the kind of misuse that should be classified as circumventing export control rules.
More broadly, the expert community argues that AI models' ability to analyze and repair software is equivalent to traditional tools such as static code analysis and fuzzing — tools that have never been subject to comparable restrictions. Critics characterize the government's decision as "hasty and dangerous," arguing that attackers can turn to alternative models while defenders are left without their most powerful tool.
A Principled Shift in AI Regulation
According to TechCrunch, the incident represents a principled shift in how the US attempts to control the spread of advanced AI. Previously, export control measures have primarily targeted semiconductors and development tools. The government now intervening directly in access to fully trained AI models marks a new and more intrusive form of regulation.
This also shines a spotlight on the enduring debate over "dual-use" technology — systems that have legitimate, valuable applications for defenders but that could in principle be exploited by attackers. The Wassenaar Arrangement, which brings together 41 countries including the US in a forum for export controls on precisely such technologies, has long struggled to strike a balance that does not inadvertently harm cyber defenders.
What Happens Next?
As of now, the situation remains unresolved. Anthropic has not published any concrete timeline for reopening the models, and the White House has not publicly responded to the expert group's demands. The case illustrates the growing tension between the government's need to control powerful AI and the security community's dependence on those same tools to protect digital infrastructure.
Source: TechCrunch, June 15, 2026