Anthropic has officially restored access to its powerful Claude Fable 5 and Claude Mythos 5 artificial intelligence models after the U.S. Department of Commerce lifted export controls that had temporarily blocked global access. The restrictions, which lasted nearly three weeks, were imposed after researchers demonstrated that the models could be manipulated into identifying software vulnerabilities and generating exploit code. Following additional security enhancements and government review, U.S. regulators determined the safeguards were sufficient to allow the models back online.
The decision marks one of the most significant AI policy reversals of the year and highlights the increasingly delicate balance between protecting national security and maintaining American leadership in artificial intelligence. It also establishes an important precedent for how the U.S. government may regulate frontier AI models capable of both defensive cybersecurity research and potential offensive misuse.
Why the AI Models Were Restricted
The export controls were introduced after Amazon researchers discovered a jailbreak technique capable of bypassing some of the models’ safety guardrails.
According to Anthropic, the prompt allowed Fable 5 to identify software vulnerabilities and, in certain cases, generate proof-of-concept exploit code. Although the company argued that similar capabilities exist in other leading AI models, U.S. officials viewed the findings as serious enough to temporarily restrict access while additional safeguards were developed.
Because Anthropic could not reliably verify the nationality of every user in real time, it suspended access globally rather than risk violating the government’s export order.
New Security Measures Cleared the Way
To satisfy federal regulators, Anthropic developed a new AI safety classifier specifically designed to detect and block the jailbreak technique that triggered the restrictions.
The company says the updated safeguard successfully prevents the exploit in more than 99% of attempts. Requests identified as potentially dangerous are automatically redirected to an older and less capable Claude model, reducing the likelihood that malicious users can weaponize the system while preserving legitimate access for most customers.
The U.S. Commerce Department’s Center for AI Standards and Innovation evaluated the new protections before recommending that export restrictions be lifted.
Fable 5 Returns While Mythos 5 Remains More Restricted
Following the government’s decision, Anthropic has begun restoring Fable 5 across:
- Claude.ai
- Claude Platform
- Claude Code
- Claude Cowork
Access to Mythos 5, however, remains more limited. The model is currently available only to a select group of vetted U.S. organizations and critical infrastructure partners while Anthropic continues working with federal agencies on additional deployment standards.
The staggered rollout reflects ongoing concerns surrounding the deployment of the company’s most capable frontier AI systems.
Anthropic Expands AI Safety Efforts
Alongside restoring access, Anthropic announced several new security initiatives designed to strengthen future AI releases.
The company has launched a HackerOne bug bounty program encouraging security researchers to report new jailbreak techniques and vulnerabilities. It is also collaborating with Amazon, Google, Microsoft
Anthropic also committed to providing U.S. government agencies with earlier access to future frontier models for safety testing before public deployment.
A New Era of AI Regulation Is Emerging
The temporary export ban represented one of the first times the U.S. government applied export controls directly to a commercially available AI model rather than physical hardware.
Instead of restricting semiconductor chips or software exports, regulators limited access to cloud-based AI services, signaling that advanced language models are increasingly being treated as strategic technologies with national security implications.
The episode has sparked broader debate across the technology industry regarding how much authority governments should have over AI deployment and who should ultimately decide when frontier models are safe enough for public release.
Competition With Global AI Rivals Continues
The temporary suspension also raised concerns about America’s competitive position in the global AI race.
Several industry leaders argued that restricting U.S.-developed models while international competitors continued advancing could unintentionally strengthen foreign AI developers, particularly those in China. Others countered that carefully balancing innovation with security will ultimately strengthen public trust and encourage responsible AI development.
The debate reflects the growing challenge governments face as AI capabilities rapidly outpace existing regulatory frameworks.
