Anthropic is restoring access to Claude Fable 5 and Mythos 5 after US authorities lifted export restrictions imposed on June 12 over concerns that a jailbreak technique could allow the models to assist with software vulnerability discovery, the company confirmed this week.
The company confirmed that both Claude Fable 5 and Mythos 5 are being redeployed after what it described as “a series of productive conversations” with the U.S. government. As part of the relaunch, Anthropic said the models will feature additional safety classifiers designed to detect and block a broader range of cybersecurity-related requests that could potentially be exploited by malicious actors.
The announcement closes a brief but highly scrutinized chapter that sparked debate across the AI industry about government intervention, model security, and how frontier AI systems should be regulated without slowing innovation.
U.S. lifts restrictions after Anthropic strengthens AI safeguards
The restoration of Claude Fable 5 follows export control measures introduced on June 12 after U.S. officials raised concerns over reports that researchers had discovered a jailbreak capable of prompting the model to identify software vulnerabilities.
To comply with the directive, Anthropic temporarily disabled access to both Claude Fable 5 and Mythos 5 for all users. In a statement issued on June 13, the company explained that the government order extended beyond foreign customers to include foreign-national employees working within Anthropic itself, leaving the company with little choice but to suspend the models entirely.
According to Anthropic, the updated versions now incorporate enhanced safety mechanisms specifically engineered to recognize and reject a wider range of cybersecurity-related prompts. The company said these additional protections directly address concerns that advanced AI systems could be manipulated through jailbreak techniques to assist with exploit development or vulnerability discovery.
U.S. Secretary of Commerce Howard Lutnick confirmed the government’s decision on X, saying officials spent the past two weeks working closely with Anthropic to review the model before approving its return under revised security requirements.
“Our goal has been to ensure these powerful AI systems can move forward while remaining aligned with U.S. security interests,” Lutnick said.
Meanwhile, White House Chief of Staff Susie Wiles emphasized that the administration remains committed to deploying advanced AI technologies “as quickly and safely as possible,” underscoring the government’s effort to balance innovation with national security considerations.
The controversy surrounding Claude Fable 5 originated after Amazon researchers reportedly demonstrated a method capable of bypassing some of the model’s safeguards, enabling it to identify software vulnerabilities under specific conditions.
Anthropic has consistently maintained that the reported exploit represented a limited or “narrow” jailbreak rather than a universal method capable of removing safety protections across the model.
The company argued that similar outputs could be produced by less capable AI systems and warned against treating isolated jailbreak findings as justification for withdrawing commercial AI models from the market.
“We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,” Anthropic said in its earlier response to government officials.
“If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
The company’s position reflects a broader debate unfolding across the AI sector. Developers increasingly acknowledge that while jailbreak attempts are inevitable, regulators must distinguish between isolated vulnerabilities and systemic failures that fundamentally compromise model safety.
Industry experts have noted that virtually every major frontier AI model has faced some form of jailbreak research, making the establishment of standardized risk classifications increasingly important as models become more capable.
Anthropic expands government partnership on AI safety
Beyond restoring access, Anthropic announced a broader expansion of its collaboration with U.S. authorities focused on AI safety, model evaluation, and misuse detection.
The company said it will provide government agencies with pre-release access to future AI models and safety systems for testing before public deployment. Anthropic also plans to increase information sharing related to jailbreak discoveries, misuse trends, and cybersecurity research while dedicating additional resources to joint safety initiatives.
The renewed partnership comes as Anthropic works alongside Amazon, Microsoft, Google, and other technology partners through Project Glasswing, a cybersecurity initiative announced in April. The collaboration aims to develop an industry-wide framework for evaluating the severity of AI jailbreaks and establishing common standards for classifying associated risks.
The company believes a consensus-based approach will help governments and AI developers respond more consistently to future security incidents while avoiding unnecessary disruption to commercial deployments.
Global policy debate over frontier AI continues
The return of Claude Fable 5 also arrives amid growing international debate over who should control access to the world’s most advanced AI systems.
On June 29, Austria urged the European Union to consider attracting Anthropic’s operations into the bloc, arguing that Europe should avoid becoming dependent on policy decisions made outside its borders.
State Secretary for Digitalization Alexander Proell said Europe risks falling behind in access to frontier AI technologies if major deployment decisions remain concentrated elsewhere.
At the same time, Anthropic has continued advocating for stronger oversight of advanced AI systems. In its June 11 policy proposal titled Policy on the AI Exponential, the company called for mandatory testing requirements, independent evaluations, stricter cybersecurity standards, and enforceable safeguards for frontier AI models capable of presenting biological, cybersecurity, or operational risks.
The restoration of Claude Fable 5 therefore represents more than the return of a single AI model. It illustrates the increasingly delicate balance governments and AI developers are attempting to strike between accelerating technological progress and ensuring that powerful AI systems remain secure, accountable, and resistant to misuse.
As governments around the world refine their approach to AI governance, the agreement between Anthropic and U.S. regulators could become an early template for how future disputes over frontier AI deployment are resolved. With Claude Fable 5 now returning under enhanced safeguards and closer government collaboration, the episode is likely to shape both regulatory policy and industry best practices for years to come.
