After nearly three weeks of export restrictions, Anthropic's Claude Fable 5 and Claude Mythos 5 models are now accessible globally following the US government lifting export controls. The company has also implemented improved cybersecurity measures and is working with industry partners to create a standard framework for assessing AI jailbreak risks.
- Global access to Claude Fable 5 restored with usage limits until July 7
- New safety classifier blocks over 99% of known bypass attempts
- Industry-wide framework for AI jailbreak assessment under development
What happened
The US government has lifted export controls that had restricted non-US citizen access to Anthropic's AI models Claude Fable 5 and Claude Mythos 5 for nearly three weeks. From July 1, the models resumed global availability across multiple platforms including the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Initial access for Pro, Max, Team, and select Enterprise users is set with 50% of previous usage limits until they transition to a usage-based credit system after July 7.
Anthropic is also working to restore service availability on major cloud providers such as AWS, Google Cloud, and Microsoft Foundry. This resumption follows a period of collaboration between the company, the US government, and partners to bolster cybersecurity protections on the models.
Why it matters
The lifted controls mark an important step in balancing AI innovation with national security and export regulations. Anthropic strengthened Fable 5 with a new safety classifier designed to detect and block bypass techniques highlighted by Amazon's security report, successfully preventing over 99% of these attempts. When flagged, requests are rerouted to a safer model version named Claude Opus 4.8, ensuring ongoing user experience while mitigating risks.
This situation highlighted the absence of standardized criteria for evaluating AI jailbreaks across the industry. Anthropic’s initiative to work with leading tech companies and government agencies on a consensus framework for assessing jailbreak severity aims to create uniformity in how AI system vulnerabilities are identified and addressed, offering a scalable response mechanism moving forward.
What to watch next
Anthropic’s continued collaboration with the US government is slated to deepen, building on prior cooperation during policy development on AI security. The company plans to implement a 24/7 monitoring team and launch a HackerOne program inviting researchers to report AI jailbreaks in Claude Fable 5, enhancing proactive threat detection and mitigation.
The development and adoption of the proposed AI jailbreak assessment framework will be a key area to monitor, as it has implications for broader industry standards and regulatory policies. How effectively Anthropic and its partners manage false positives in security classifiers while maintaining model usability will also be crucial for user satisfaction and trust.