The disruption began on June 12, 2026, when the Trump administration issued an emergency export control directive forcing Anthropic to pull its high-capability systems from service. The enforcement action followed weeks of rising concern after security researchers at Amazon documented a vulnerability where the underlying frontier models could be "jailbroken" to identify zero-day software vulnerabilities and assist in generating exploitation code.

Because Anthropic lacked a real-time automated nationality verification infrastructure to selectively block foreign access at the API level, the company chose to institute a total global service suspension to ensure compliance with federal law.

Following intensive negotiations and technical adjustments with American federal authorities, U.S. Commerce Secretary Howard Lutnick confirmed that the export controls were officially lifted on June 30, 2026. Anthropic achieved clearance by rolling out a specialized, real-time automated safety classifier layer designed to proactively intercept malicious code-generation and cyber-attack staging. Global access to the ecosystem officially went back online on July 1, 2026.

Launch of Claude Sonnet 5

Coinciding with the system restoration on June 30, Anthropic also officially launched Claude Sonnet 5. Positioned as a safer, highly accessible baseline designed for general use, Sonnet 5 introduces advanced autonomous capabilities with real-time cybersecurity safeguards enabled by default. The model is engineered to handle complex, multi-step software engineering work, including terminal operations and browser-based agentic workflows, at a fraction of the cost of flagship systems.

As part of the redeployment, Anthropic clarified the distinct operational roles and security controls governing its tier-five architectures.

Claude Fable 5 serves as the company's flagship, widely released frontier model built for demanding reasoning and long-horizon agentic work. It is now globally available to public and enterprise developers across the Claude Platform with the newly mandated safety classifiers fully active.

Sharing the same raw capability as Fable 5, Claude Mythos 5 is a specialized version that completely lacks the standard safety classifiers, optimizing it for advanced vulnerability discovery and biodefense screening. Mythos 5 remains under strict federal oversight and is strictly restricted to deeply vetted U.S. critical infrastructure organizations through Anthropic’s Project Glasswing.

The developer friction trade-off

A company spokesperson emphasized that the updated rollout reflects "a continued commitment to responsible deployment and strengthened safeguards across its AI systems," noting the addition of enhanced monitoring tools and filtering systems to improve model reliability.

However, industry analysts note that the rapid resolution highlights the growing tension between rapid AI development and stringent regulatory oversight. To satisfy the U.S. government mandates, Anthropic's new automated safety classifier operates with a highly conservative safety margin. Early developer feedback indicates that while it successfully blocks malicious prompts, it frequently triggers "false positives"—inadvertently flagging benign requests during routine application coding and software debugging.

To keep developer workflows from crashing, Anthropic has integrated a fallback protocol that automatically down-routes a blocked request to be processed by the older, more permissive Claude Opus 4.8 architecture, illustrating the delicate trade-offs required to enforce watertight regulatory compliance on frontier AI.