Anthropic Gets Partial US Relief on Mythos 5 Restrictions

Anthropic said it is restoring access to Mythos 5 for a set of US organizations that "operate and defend critical infrastructure."

The company said in a statement that the government had granted a limited carveout from the restrictions imposed on its latest AI models. The carveout applies to entities involved in power grids, water systems, transportation networks, and other essential services.

Anthropic did not disclose the exact number of organizations covered or the specific terms of the arrangement. The company said the approval followed a review process that assessed the model's safety protocols and the operational needs of critical infrastructure operators.

The restrictions on Mythos 5 were put in place earlier this year, citing concerns about the model's advanced capabilities and potential misuse. Anthropic had argued that the restrictions were too broad and hampered legitimate uses in security and infrastructure protection.

The partial relief does not apply to commercial or consumer applications of Mythos 5. Those remain subject to the original restrictions. Anthropic said it continues to work with regulators on a broader framework for model deployment.

The carveout covers organizations that have existing security clearances and meet specific operational criteria, the company said. It did not name any of the organizations that would gain access.

Anthropic's statement emphasized that the restored access is limited to defensive and protective functions, not offensive or surveillance-related uses. The company said it will monitor compliance through its existing oversight mechanisms.

The decision comes as the US government weighs broader rules for advanced AI models, balancing national security concerns with the needs of critical infrastructure operators. No timeline has been set for a final regulatory framework.