After a temporary suspension prompted by a US Department of Commerce directive, Anthropic will restore full global access to its Claude Fable 5 AI model starting July 1, implementing enhanced safety measures to address prior vulnerabilities.
- Access to Claude Fable 5 resumes globally on July 1 with new security filters.
- Suspension followed US export controls triggered by vulnerability disclosures.
- Anthropic collaborates with tech giants on shared standards for AI jailbreak prevention.
What happened
On June 12, the US government issued a directive requiring Anthropic to block foreign nationals from using its Claude Fable 5 and Mythos 5 AI models due to concerns over enabling cybersecurity vulnerabilities. Unable to verify user nationalities in real time, Anthropic suspended access to both models entirely. This move came after researchers at Amazon revealed exploits that could bypass Fable 5’s safeguards and make it generate code demonstrating software flaws.
Following productive discussions with US authorities, the Department of Commerce lifted the export controls, allowing Anthropic to reinstate Claude Fable 5 access globally as of July 1. Mythos 5 had already been reintroduced earlier to a limited set of approved US organizations, with expanded access authorized beginning June 26. Anthropic incorporated a new safety classifier to block the cybersecurity risks flagged, reportedly preventing problematic outputs in more than 99% of attempts.
Why it matters
This episode highlights the intricate balance between AI innovation and regulatory compliance in sensitive areas such as cybersecurity. It underscores the challenges companies face in deploying advanced AI globally amid export control regimes and evolving risk assessments. The initial suspension disrupted users around the world, emphasizing the importance of real-time nationality verification and robust security frameworks in generative AI services.
Anthropic’s proactive collaboration with the government, industry leaders like Amazon, Microsoft, and Google, and the development of shared standards for grading AI jailbreak severity represent significant steps toward safer AI deployment. This case will likely influence how AI providers approach risk management and compliance monitoring, shaping future regulatory and operational strategies in the field.
What to watch next
Monitoring the performance and reliability of the new safety classifier implemented by Anthropic across Claude Fable 5 will be key to ensuring the cybersecurity risks remain mitigated. Users and regulators will be attentive to potential attempts to circumvent the updated safeguards and the company’s responses to such challenges.
The ongoing rollout of Claude Fable 5 access on major cloud platforms like AWS, Google Cloud, and Microsoft Foundry will broaden availability and create new operational dynamics. Additionally, how the AI industry standard for grading jailbreak severity evolves, supported by major tech stakeholders, will shape future AI safety regulations and the governance landscape.