Anthropic Fable 5 May Return After Weeks Under U.S. Restrictions

Anthropic’s advanced Fable 5 artificial intelligence model could return to general availability as early as this week after spending more than two weeks offline because of U.S. government security restrictions. Reports from Axios and Reuters indicate that officials may soon lift the limits placed on the model, although no final decision has been confirmed.

The expected move follows a partial easing of restrictions on Claude Mythos 5, Fable 5’s sibling model. Axios reported that the U.S. Commerce Department has approved limited access for authorized users, while Reuters said Anthropic has begun redeploying Mythos 5 to selected U.S. organizations responsible for protecting critical infrastructure. Developers have been waiting for Fable 5 to return so they can resume coding agents, large-scale software projects and long-running development workflows that were interrupted by the suspension.

Anthropic introduced Fable 5 as a Mythos-class AI model designed for broad public use while incorporating stronger safeguards for higher-risk areas such as cybersecurity and biology. The company said it shares the same underlying technology as Mythos 5 but includes additional protections. Fable 5 also delivers major improvements in software engineering, coding, vision, memory, knowledge work and long-duration reasoning tasks, making it one of the industry’s most capable AI systems.

Shortly after its launch, users widely shared examples of Fable 5’s performance across social media and online forums, highlighting its ability to maintain long context windows, complete complex code refactoring and perform advanced reasoning with fewer interruptions than competing models. Despite its popularity, Anthropic abruptly suspended access after announcing that a U.S. export-control directive required the company to block Fable 5 and Mythos 5 for foreign nationals, including foreign-national employees within Anthropic. To comply with the directive, the company disabled the models for all customers.

Before the suspension, Anthropic said Fable 5 became generally available on June 9 through the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud and Microsoft Foundry. Mythos 5, however, remained restricted to organizations participating in the company’s pre-approved Project Glasswing program.

The government’s concerns stem from the dual-use nature of advanced AI systems. Anthropic says Mythos 5 significantly improves performance in cybersecurity, biology and healthcare benchmarks. While those capabilities can help security teams identify vulnerabilities and strengthen defenses, they can also be used to enhance offensive cyber operations. That balance between innovation and national security has shaped the restrictions placed on both models.

Anthropic recently confirmed that Mythos 5 has been approved again for deployment to select U.S. organizations that operate and defend critical infrastructure. Reuters reported that more than 100 companies and institutions, including numerous Fortune 500 firms, are expected to receive access under the updated policy.

The pause also created opportunities for competitors. On June 26, OpenAI introduced preview versions of GPT-5.6 Sol, Terra and Luna, describing Sol as its most capable cybersecurity model to date and highlighting improvements in coding, defensive testing and long-horizon security tasks. OpenAI said access would remain limited during the preview before a broader release.

Chinese AI developers also gained attention during Anthropic’s absence. Z.ai released the open-weight GLM-5.2 model shortly after Fable 5 became unavailable. Reuters reported that benchmark testing placed GLM-5.2 close to leading closed-source models, while Z.ai said it features a one-million-token context window and stronger coding performance. DeepSeek also expanded its AI offerings with its V4 Preview model, introducing a one-million-token context window and sparse attention designed to reduce computing and memory requirements. Reuters further reported that DeepSeek-V4 has been adapted for Huawei Ascend processors as China continues reducing reliance on foreign AI hardware.

The growing availability of powerful foreign AI models has complicated the policy debate in the United States. Restricting domestic frontier AI systems may reduce immediate security risks, but it could also encourage developers to adopt overseas alternatives that are less restricted, open-weight or easier to self-host. Axios noted that open-source models such as GLM-5.2 have also raised cybersecurity concerns because they can be downloaded, modified and potentially stripped of built-in safeguards.

If Fable 5 returns in the coming days, Anthropic could regain some of the momentum it lost during the outage. However, the company will also face questions from customers about model availability, access policies, suspension risks and long-term reliability as competition in the frontier AI market continues to intensify.