Trump Admin releases Anthropic Mythos to be used by more than 100 US companies, agencies

Trump Administration Eases Ban on Anthropic's Mythos 5

The Trump administration is partially reversing its recent ban on Anthropic's powerful cybersecurity AI model, Mythos 5. Just two weeks after the model was pulled from the market over security concerns, the government is now permitting its use by a select group of more than 100 U.S. companies and federal agencies. This move signals a significant adjustment in the administration's approach to regulating high-capability AI systems, allowing trusted partners to access advanced tools while a broader ban remains in effect.

Key Details of the Reversal

• Model Affected: Access is being restored for Claude Mythos 5, Anthropic's most advanced cybersecurity model.
• Scope of Access: Over 100 specified U.S. government agencies and companies defending critical infrastructure.
• Personnel: Access is notably granted to both American and non-American employees at the approved organizations.
• Excluded Model: The directive did not address the status of Fable 5, a related model that had been more widely released before the ban.

The policy shift was detailed in a letter from Commerce Secretary Howard Lutnick to Anthropic, which stated that "appropriate safeguards are in place to permit certain trusted partners" to use the model. In a public statement on X, Anthropic confirmed the development, noting it was working to restore access quickly for these organizations and continuing discussions with the government to expand availability. The original ban was enacted after security researchers reportedly bypassed the safety guardrails on both Mythos 5 and Fable 5.

A New Regulatory Playbook

This decision suggests the administration is forging a more nuanced regulatory framework, creating a distinction between general public access and controlled deployment within secure, vetted environments. By approving Mythos 5 for a specific list of users while keeping the more broadly released Fable 5 off the market, regulators are implementing a tiered approach to risk management. This selective approval could serve as a template for governing other specialized AI systems, balancing the operational need for advanced tools against potential national security threats.

The administration's selective approval of Mythos 5 for trusted partners, while leaving Fable 5 sidelined, establishes a clear precedent for a two-tiered regulatory system for powerful AI: one for vetted government and enterprise use, and a far more restrictive one for public-facing models.