US government removes restrictions on Fable 5 and Mythos 5, freeing Anthropic AI models with new protections

Anthropic, a leading artificial intelligence developer, has reactivated access to its advanced AI models, Fable 5 and Mythos 5. The move comes after the United States government revoked export controls that had suddenly blocked access earlier this month due to national security concerns.
The Fable 5 template will be available to users worldwide starting July 1st. The company confirmed that the release includes its platforms, such as Claude.ai and Claude Cowork, as announced in an official post.
For Mythos 5, access has been reinstated to select organizations in the United States. Government approval for this specific model was granted on June 26th.
Anthropic is currently coordinating with North American authorities to expand the reach of Mythos 5. The objective is to make the model available to a greater number of national and international partners, through its initiative called Project Glasswing.
The company disclosed that, “after a series of productive conversations with the US government, we are redeploying the (Fable 5) model with a new set of classifiers to identify and block more cybersecurity tasks.” This action reflects a joint effort to strengthen the resilience of AI systems.

In a post on Platform The collaboration aims to create security standards for the industry.
The Indian market represents the second largest for Anthropic globally. Recently, the AI company entered into an agreement with TCS to train 50,000 of its employees using Anthropic models.
Additionally, a partnership with Infosys was announced, with a focus on implementing advanced enterprise AI solutions, which will make use of the Claude family of models.
Sources close to the matter indicated that some government agencies and private companies in India have already had access to the Mythos model. This occurred in the context of a cybersecurity initiative linked to Project Glasswing.
The US’ initial decision to impose restrictions came after Amazon researchers discovered a vulnerability known as a “jailbreak”. This method managed to bypass Fable 5’s security safeguards, allowing the team to identify flaws in the software and demonstrate how to exploit them.
To address the issue, Anthropic collaborated directly with the US government to train an “enhanced security classifier.” This system is capable of identifying and blocking the specific bypass technique in more than 99% of cases, reinforcing the integrity of the model.
Anthropic reinforced its commitment to expanding collaboration with the US government for safeguards testing and development. This encompasses providing early access to models for evaluation, sharing information about intrusions and misuse, and earmarking resources for joint research into AI security.






