Anthropic has restored global access to its Claude Fable 5 AI model after the U.S. Department of Commerce removed the export controls that had been introduced in mid-June. The restrictions were lifted on June 30, allowing the company to begin bringing the model back online from July 1. Claude Fable 5 is now being restored across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. The decision ends a temporary shutdown that affected users around the world.
The export controls were originally introduced after concerns were raised about a jailbreak technique that could bypass the model’s built-in safety protections. Researchers from Amazon reported that the method allowed the AI model to identify software vulnerabilities and, in one instance, generate code that demonstrated how a flaw could be exploited. Because the government order applied to all foreign nationals, including Anthropic’s own non-U.S. employees, the company decided to disable the model for everyone instead of trying to verify each user’s nationality in real time.
Anthropic maintained that the reported jailbreak did not reveal any unique or hidden capability in Claude Fable 5. According to the company, similar requests could also be completed by several other publicly available AI models, including its own Claude Opus 4.8 as well as competing models from other AI developers. The company said the model’s responses mainly involved identifying known software issues that are commonly used in defensive cybersecurity work rather than creating new offensive capabilities.
To address the government’s concerns, Anthropic developed a new safety classifier specifically designed to detect and block the jailbreak method that triggered the restrictions. The company stated that the updated protection blocks the reported technique in more than 99% of attempts. If the system detects this specific jailbreak, the request is automatically redirected to the less powerful Claude Opus 4.8 model, and the user is informed about the change. Anthropic acknowledged that the new protection may occasionally block some legitimate coding or debugging requests.
While Claude Fable 5 has returned to users worldwide, Anthropic’s more advanced Mythos 5 model remains under tighter controls. Access to Mythos 5 was partially restored earlier for a limited group of around 100 U.S. companies and federal agencies involved in protecting critical infrastructure. Anthropic said it is continuing to work with U.S. authorities to expand access in a controlled and secure manner while ensuring that appropriate safeguards remain in place.
The incident has sparked wider discussions about how governments should regulate advanced AI models without slowing innovation. Some researchers and industry experts believe the original restrictions were too broad, especially since similar capabilities exist in other AI systems. At the same time, officials argued that advanced AI models capable of assisting with cybersecurity tasks require stronger oversight to reduce the risk of misuse through jailbreak techniques or other safety bypass methods.
Anthropic also announced several additional security measures following the restoration of Claude Fable 5. The company plans to respond much faster to future high-risk jailbreak reports and is creating a dedicated team to monitor new attempts around the clock. It has also launched a HackerOne bug bounty program, encouraging security researchers to responsibly report new jailbreak methods so they can be fixed before they are widely abused.
The company said it will continue working closely with the U.S. government on future AI safety standards and will allow officials to evaluate upcoming frontier AI models before public release. Anthropic believes this approach can improve security while keeping advanced AI available for developers, businesses, and researchers. The restoration of Claude Fable 5 marks an important step in balancing AI innovation with national security concerns and stronger protections against misuse.
Stay alert, and keep your security measures updated!
Source: Follow cybersecurity88 on X and LinkedIn for the latest cybersecurity news
