Anthropic Reintroduces Claude Fable 5 Globally
Anthropic has announced the global reactivation of its AI model, Claude Fable 5, following the removal of export restrictions by the U.S. Commerce Department on June 30. These controls, initially imposed two and a half weeks prior, were lifted, allowing the model to be accessible again from July 1 on various platforms including Claude.ai and the Claude Platform.
Understanding the Export Restrictions
The export controls initially enforced by the U.S. government aimed to restrict the usage of Claude Fable 5 and its counterpart, Mythos 5, by foreign nationals. The order, effective from June 12, mandated Anthropic to halt both models for non-citizens, even affecting its own international staff, due to the challenge of verifying users’ nationalities in real-time.
The restrictions were triggered by a ‘jailbreak’—a prompt enabling the model to bypass its safety protocols. Researchers from Amazon discovered this vulnerability in Fable 5, which could potentially expose software flaws and suggest exploitative code. Despite Anthropic’s assertion that such vulnerabilities exist in other models like its own Claude Opus 4.8, the incident prompted emergency measures.
Addressing Security Concerns
In response to security concerns, Anthropic developed a new safety filter, known as a classifier, to detect and block the specific technique identified in the report. The company claims this filter now successfully prevents the issue in over 99% of attempts, redirecting blocked requests to the less advanced Opus 4.8 model.
While Claude Fable 5 resumes widespread availability, Mythos 5 remains under tighter controls, with access granted only to select U.S. enterprises and government agencies as of June 26. The company continues to collaborate with government bodies to expand access safely.
Negotiations and Industry Impact
Commerce Secretary Howard Lutnick confirmed that the decision to lift the restrictions resulted from extensive discussions with Anthropic. The company committed to ongoing security assessments and coordination for future AI model launches.
Reports suggested that Amazon’s initial concerns and CEO Andy Jassy’s input influenced the original export ban. Some industry analysts viewed the reversal as a correction of an overreach, while others highlighted the competitive advantage gained by foreign models during the pause.
Anthropic is advocating for a standardized approach to assessing the risks associated with AI model jailbreaks, collaborating with major tech companies like Amazon and Google to establish safety metrics. The company is also initiating a HackerOne program for reporting new vulnerabilities and offering U.S. authorities early access to test future models.
As AI technology continues to evolve, the challenge remains to balance innovation with security, ensuring that advanced models contribute positively to society while mitigating potential risks.
