On June 30, 2026, Anthropic announced the gradual lifting of access restrictions for its AI models, "Claude Fable 5" and "Claude Mythos 5," following the removal of U.S. government export regulations. Starting July 1, Claude Fable 5 will be available to global users via the Claude Platform and Claude.ai. Access through AWS, Google Cloud, and Microsoft Foundry is also scheduled for phased reactivation.

The service suspension began on June 12 due to U.S. export controls. Anthropic initially suspended access for all customers, as it lacked a reliable real-time method to verify the nationality of users, a requirement under the regulation. Claude Fable 5 is now resuming global service, with plans featuring the model limited to 50% of the weekly usage cap until July 7. Conversely, Claude Mythos 5 has only been restored for select U.S. organizations approved by the government on June 26.

The suspension followed reports from Amazon researchers who demonstrated that Claude Fable 5's safeguards could be bypassed to identify software vulnerabilities and generate exploit code. Over the past two weeks, Anthropic worked with the U.S. government and Amazon to conclude that these methods represented edge cases in defensive cybersecurity work. In response, the company trained a new safety classifier capable of blocking such techniques in over 99% of cases; blocked requests are now routed to Claude Opus 4.8.

Following this incident, the company has proposed an industry-standard framework to evaluate the severity of AI jailbreaks. Working alongside Amazon, Google, and Microsoft, Anthropic aims to define criteria based on attack ease and capability impact. Additionally, the company is transitioning to a system requiring pre-assessment and joint research with the U.S. government for advanced AI models involved in national security.

On June 30, 2026, Anthropic announced the gradual lifting of access restrictions for its AI models, "Claude Fable 5" and "Claude Mythos 5," following the removal of U.S. government export regulations. Starting July 1, Claude Fable 5 will be available to global users via the Claude Platform and Claude.ai. Access through AWS, Google Cloud, and Microsoft Foundry is also scheduled for phased reactivation.

The service suspension began on June 12 due to U.S. export controls. Anthropic initially suspended access for all customers, as it lacked a reliable real-time method to verify the nationality of users, a requirement under the regulation. Claude Fable 5 is now resuming global service, with plans featuring the model limited to 50% of the weekly usage cap until July 7. Conversely, Claude Mythos 5 has only been restored for select U.S. organizations approved by the government on June 26.

The suspension followed reports from Amazon researchers who demonstrated that Claude Fable 5's safeguards could be bypassed to identify software vulnerabilities and generate exploit code. Over the past two weeks, Anthropic worked with the U.S. government and Amazon to conclude that these methods represented edge cases in defensive cybersecurity work. In response, the company trained a new safety classifier capable of blocking such techniques in over 99% of cases; blocked requests are now routed to Claude Opus 4.8.

Following this incident, the company has proposed an industry-standard framework to evaluate the severity of AI jailbreaks. Working alongside Amazon, Google, and Microsoft, Anthropic aims to define criteria based on attack ease and capability impact. Additionally, the company is transitioning to a system requiring pre-assessment and joint research with the U.S. government for advanced AI models involved in national security.