Anthropic has announced the U.S. Commerce Department has lifted export controls on its advanced Claude Fable 5 and Mythos 5 AI models. The decision comes less than three weeks after the company was ordered to suspend access due to national security concerns. This rapid reversal follows intensive collaboration between the AI firm and government officials to address the identified risks.
A Swift Reversal on Export Controls
The company abruptly suspended access to both models for all users on June 12 after the government order took immediate effect. Anthropic stated it had no reliable way to verify user nationality in real-time, necessitating a complete shutdown. Following the resolution, the company confirmed it would begin restoring global access to Fable 5 on July 1.
U.S. Commerce Secretary Howard Lutnick confirmed the collaborative effort, stating officials worked closely with Anthropic to analyze and approve the model. While Fable 5 is being made widely available, access to the more powerful Mythos 5 remains limited to a select group of trusted U.S. organizations. This phased rollout reflects a cautious approach to managing the technology's capabilities.
The Catalyst for the Ban
The initial export control directive was triggered by a report from Amazon researchers who discovered a method to bypass Fable 5's safeguards. This "jailbreak" allowed the model to identify software vulnerabilities and, in one instance, generate code to exploit one. The finding immediately raised alarms within the government regarding the model's potential for misuse in cyberattacks.
In its defense, Anthropic conducted tests showing that many less capable models, including those from competitors, could replicate the reported behavior. The company argued that the technique did not expose any unique, high-risk capabilities exclusive to its frontier models. This evidence was a crucial part of its discussions with regulators to demonstrate the context of the vulnerability.
To resolve the government's concerns, Anthropic developed an improved safety classifier that specifically targets and blocks the bypass method. This new safeguard, which is effective in over 99% of cases, was reviewed and validated by the Commerce Department’s Center for AI Standards and Innovation. The implementation of this technical fix was instrumental in the decision to lift the export controls.
Broader Implications for the AI Industry
This incident highlights the increasing oversight Washington is applying to the release of powerful new AI systems. The government's active vetting of which companies can access advanced models has drawn some criticism from industry leaders for potentially stifling innovation. The episode underscores the delicate balance between fostering technological leadership and mitigating potential national security threats.
In response to these events, Anthropic is championing the creation of a shared industry framework for assessing AI jailbreaks. The company is partnering with Amazon, Microsoft, and Google to develop a consistent standard for judging the severity of model vulnerabilities. This initiative aims to help developers triage new findings and communicate risks more effectively to government partners.
The AI firm has also committed to deepening its collaboration with the U.S. government on frontier AI security. This includes providing pre-release model access for evaluation, sharing threat intelligence rapidly, and dedicating resources for joint research. These steps represent a proactive effort to build a more durable and transparent process for deploying powerful AI safely.
The swift resolution of the export controls on Anthropic's models marks a significant moment in the relationship between AI developers and regulators. While access has been restored, the event has accelerated the push for industry-wide safety standards and transparent government oversight. The outcome suggests a future built on close collaboration to manage the risks and unlock the benefits of advanced artificial intelligence.