Blab AI
Anthropic Restores Fable 5 After U.S. Lifted Export Controls
On July 1, Anthropic will lift the curtain on its Fable 5 model after the U.S. Department of Commerce lifted export controls that had halted access a month earlier.
The pause began on June 12, when the company was instructed to block all foreign access to Fable 5 and its sibling Mythos 5. The directive followed a jailbreak technique uncovered by researchers that allowed the models to bypass built‑in safety filters. By exploiting the loophole, the system could identify security vulnerabilities in third‑party software—an ability that raised concerns about potential misuse.
Anthropic’s investigation revealed that similar vulnerabilities could be exposed by less capable models from other AI firms. In response, the company increased the overall safety margin of Fable 5 and added new safeguards designed to reduce the likelihood that a jailbreak could be used to generate harmful content. The firm stressed that no AI system can be made completely resistant to jailbreaks and that the industry currently lacks a common framework for responding to new jailbreak methods.
“We are working with government partners on initiatives that could eventually serve as the foundation for systematic industry‑wide rules,” the company said. The earlier export‑control directive was issued after the White House voiced concerns that Mythos, a model capable of autonomously identifying software vulnerabilities, could facilitate cyberattacks. In May, the Wall Street Journal reported that the Treasury Secretary met with executives from major banks to discuss the risks. Anthropic had not released Mythos to the public; it was made available only to a limited group of partners. Instead, the company introduced Fable for public use, describing it as the same underlying model with significantly stronger safety restrictions.
CEO Dario Amodei said the company had implemented robust safety safeguards and had conducted extensive testing in cooperation with U.S. authorities before the planned public release. The firm also noted that perfect protection against jailbreaks is impossible, and its objective has been to make jailbreak methods either highly specialized or prohibitively expensive to develop.
The U.S. government lifted the export controls on June 30, clearing the way for Fable 5 and Mythos 5 to be made available again. According to a statement from the Department of Commerce, the decision followed a review of the models’ safety measures and a discussion with Anthropic. The company confirmed that it would begin restoring access to Fable 5 on July 1, with a classifier that blocks jailbreak attempts in more than 99 % of cases.
The incident highlights the growing tension between rapid AI development and the need for robust safety mechanisms. While Anthropic has taken steps to strengthen its models, the company acknowledges that the broader industry still faces challenges in coordinating responses to emerging jailbreak techniques. The lifting of export controls also signals a shift in U.S. policy toward allowing advanced AI models to be used more widely, provided that safety and security concerns are addressed.
At present, Fable 5 is available to users in the United States and other countries that are not subject to the export restrictions. Anthropic is continuing to monitor the model’s performance and to collaborate with regulators on best practices for AI safety. The company has not announced plans to release Mythos to the public, and it remains unclear whether additional safeguards will be added to the model in the future. The situation underscores the need for ongoing dialogue between AI developers, governments, and industry stakeholders to ensure that powerful language models can be deployed safely and responsibly.
The pause began on June 12, when the company was instructed to block all foreign access to Fable 5 and its sibling Mythos 5. The directive followed a jailbreak technique uncovered by researchers that allowed the models to bypass built‑in safety filters. By exploiting the loophole, the system could identify security vulnerabilities in third‑party software—an ability that raised concerns about potential misuse.
Anthropic’s investigation revealed that similar vulnerabilities could be exposed by less capable models from other AI firms. In response, the company increased the overall safety margin of Fable 5 and added new safeguards designed to reduce the likelihood that a jailbreak could be used to generate harmful content. The firm stressed that no AI system can be made completely resistant to jailbreaks and that the industry currently lacks a common framework for responding to new jailbreak methods.
“We are working with government partners on initiatives that could eventually serve as the foundation for systematic industry‑wide rules,” the company said. The earlier export‑control directive was issued after the White House voiced concerns that Mythos, a model capable of autonomously identifying software vulnerabilities, could facilitate cyberattacks. In May, the Wall Street Journal reported that the Treasury Secretary met with executives from major banks to discuss the risks. Anthropic had not released Mythos to the public; it was made available only to a limited group of partners. Instead, the company introduced Fable for public use, describing it as the same underlying model with significantly stronger safety restrictions.
CEO Dario Amodei said the company had implemented robust safety safeguards and had conducted extensive testing in cooperation with U.S. authorities before the planned public release. The firm also noted that perfect protection against jailbreaks is impossible, and its objective has been to make jailbreak methods either highly specialized or prohibitively expensive to develop.
The U.S. government lifted the export controls on June 30, clearing the way for Fable 5 and Mythos 5 to be made available again. According to a statement from the Department of Commerce, the decision followed a review of the models’ safety measures and a discussion with Anthropic. The company confirmed that it would begin restoring access to Fable 5 on July 1, with a classifier that blocks jailbreak attempts in more than 99 % of cases.
The incident highlights the growing tension between rapid AI development and the need for robust safety mechanisms. While Anthropic has taken steps to strengthen its models, the company acknowledges that the broader industry still faces challenges in coordinating responses to emerging jailbreak techniques. The lifting of export controls also signals a shift in U.S. policy toward allowing advanced AI models to be used more widely, provided that safety and security concerns are addressed.
At present, Fable 5 is available to users in the United States and other countries that are not subject to the export restrictions. Anthropic is continuing to monitor the model’s performance and to collaborate with regulators on best practices for AI safety. The company has not announced plans to release Mythos to the public, and it remains unclear whether additional safeguards will be added to the model in the future. The situation underscores the need for ongoing dialogue between AI developers, governments, and industry stakeholders to ensure that powerful language models can be deployed safely and responsibly.
