Anthropic’s flagship AI models, Claude Fable 5 and Mythos 5, have been offline for roughly two and a half weeks after the US government intervened with an export control order citing national security risks. The shutdown, which began around June 12-13, marks one of the most dramatic collisions between federal authority and the AI industry to date.

What actually happened

The Trump administration’s concerns centered on vulnerabilities in both models that could potentially be exploited through “jailbreaking,” a technique where users manipulate an AI system into bypassing its built-in safety guardrails.

Anthropic, for its part, has described the situation as a potential misunderstanding. The company expressed regret for the disruption to users who suddenly found themselves locked out of tools they’d come to rely on.

By late June, the government relented slightly. Mythos 5 was partially reinstated for a select group of approved US cybersecurity professionals. Fable 5, however, remains under stricter limitations with no clear timeline for broader restoration.

The broader context

Fable 5 and Mythos 5 represent some of Anthropic’s most advanced work in both AI performance and safety. The company built them with state-of-the-art safety features, which makes the government’s intervention all the more striking.

The incident also arrives amid a broader push by the Trump administration to tighten export controls on advanced technology. The US Commerce Department issued the directive, blocking foreign nationals from accessing the models. Applying that same framework to cloud-hosted AI services represents a meaningful expansion of how export controls function in practice.

What this means for investors and the AI sector

For investors, the immediate concern is regulatory unpredictability. Anthropic is one of the most well-funded AI companies on the planet, backed by billions in investment from major players including Amazon. When a company of that scale gets its flagship products pulled offline by a government order, it forces a recalibration of risk across the entire sector.

While Anthropic’s most powerful models sit idle, rivals like OpenAI, Google DeepMind, and others continue operating.

The selective reopening of Mythos 5 to approved cybersecurity professionals offers a possible template for how this plays out going forward. Rather than blanket access or blanket restrictions, the government may push toward a tiered model where different users get different levels of access based on vetting and use case.