Anthropic Restores Claude Fable 5 Access Globally After US Lifts Export Controls
Anthropic has restored worldwide access to its Claude Fable 5 AI model following the U.S. Commerce Department's decision to lift export controls imposed due to concerns over potential misuse, particularly jailbreaking.
Anthropic has successfully restored global access to its Claude Fable 5 AI model, marking the end of U.S. Commerce Department export controls that were implemented approximately two and a half weeks prior. The lifting of these controls, effective June 30, allows Fable 5 to be available once again across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork starting July 1.
These export controls, enacted on June 12, had restricted the use of Fable 5 and its more tightly controlled counterpart, Mythos 5, by any foreign national, both within and outside the United States. Due to the immediate effect of the order and the practical difficulties in verifying user nationality in real-time, Anthropic had opted to shut down access to both models for all users.
The primary catalyst for the government's intervention was the discovery of a "jailbreak" – a specific prompt that enabled the AI model to bypass its safety protocols. Researchers at Amazon identified a prompt within Fable 5 that caused the model to flag several software flaws and, in one instance, generate code demonstrating how a vulnerability could be exploited. Anthropic, however, downplayed the finding, asserting that similar prompts could bypass safety measures in other advanced models, including its own Claude Opus 4.8 and OpenAI's GPT-5.5.
To address the government's concerns, Anthropic developed and implemented a new safety filter, termed a classifier, specifically designed to detect and block the identified jailbreaking technique. The company reports that this new filter successfully prevents the technique in over 99% of attempts, redirecting blocked requests to the less powerful Opus 4.8 model and informing the user. This enhanced security measure, however, may lead to an increase in false positives for legitimate coding and debugging tasks.
Meanwhile, Mythos 5, the underlying model with fewer safety guardrails, remains under stricter limitations. Access was reinstated on June 26 for a select group of approximately 100 U.S. companies and federal agencies involved in critical infrastructure protection. Anthropic continues to collaborate with the government to explore avenues for broader access to this model.
Commerce Secretary Howard Lutnick confirmed the reversal, stating that the department had engaged in a two-week review with Anthropic. As part of the agreement, Anthropic has committed to proactively searching for security vulnerabilities, coordinating on future model launches, and reporting any instances of malicious use. This resolution highlights a complex interplay between AI development, national security, and international trade, as the U.S. grapples with regulating powerful AI technologies.
The situation also underscores the competitive landscape of AI development, with concerns that such controls could cede ground to rapidly advancing Chinese open-source models. Anthropic is also proposing a standardized method for ranking the severity of AI jailbreaks, collaborating with major tech players like Microsoft and Google. This initiative aims to provide a more consistent framework for assessing risks, with a focus on capability gain, breadth of attack, ease of weaponization, and discoverability.
This resolution of the Fable 5 export controls signifies a temporary détente in the ongoing debate surrounding AI safety and regulation. It reflects an improvised approach by the U.S. government to manage the rapid advancement of frontier AI models, as mandatory licensing or review processes are still under development. The incident also prompts broader industry discussions on responsible AI deployment and the challenges of balancing innovation with security.