The US Department of Commerce ordered Anthropic to suspend access to its Claude Fable 5 and Claude Mythos 5 AI models on June 12, just three days after both went live. The directive applies to all foreign nationals, including Anthropic’s own employees, effectively killing public and enterprise access in one swift move.

What happened and why it matters

Anthropic launched both models on June 9 as part of its latest generation of frontier AI systems. Mythos 5, the more powerful of the two, sits at the center of Anthropic’s Project Glasswing, a collaborative effort involving AWS, Microsoft, and CrowdStrike. The model has demonstrated autonomous vulnerability discovery capabilities, which is a polite way of saying it can find security holes in software without human guidance.

The problem that apparently spooked regulators involves a jailbreak vulnerability in Fable 5. Exploiting it could reportedly unlock the unrestricted capabilities of Mythos 5.

The export control directive was reportedly influenced by discussions at the White House level, with Amazon CEO Andy Jassy among those flagging concerns. Amazon is a major investor in Anthropic and provides cloud infrastructure through AWS.

Anthropic has argued that the risks posed by Fable 5 and Mythos 5 are comparable to those from other frontier AI models already on the market. Dario Amodei, Anthropic’s CEO, has called for mandatory third-party testing of AI models for cybersecurity and safety risks.

The DeFi security angle

DeFi protocols have already faced over $840 million in hacks in 2026. Project Glasswing’s partnership with CrowdStrike signals that Anthropic is well aware of the dual-use problem between defensive and offensive AI capabilities.

For DeFi protocols that were beginning to integrate AI-powered security tools, the suspension creates an immediate gap. Any project relying on Anthropic’s latest models for threat analysis or code auditing lost access without warning. The move also raises questions about the reliability of building critical security infrastructure on top of tools that can be switched off by government order in 72 hours.

A precedent with long shadows

The use of export controls to restrict access to an AI model is a significant escalation in how the US government regulates artificial intelligence. Export controls have traditionally been the tool of choice for restricting semiconductor sales to China or limiting access to nuclear technology. Applying them to a software model, one that was publicly available for three days, is a different beast entirely.

What this means for investors

Anyone with exposure to Anthropic, whether directly or through Amazon’s significant stake, is watching a company whose flagship products just got grounded by regulatory intervention.

The competitive landscape shifts too. OpenAI, Google DeepMind, and other frontier labs now have a window where Anthropic’s most capable models are offline. If Anthropic can patch the jailbreak vulnerability and demonstrate compliance with whatever security benchmarks the Commerce Department sets, the models could return.