Anthropic’s Mythos AI Model Faces Unauthorized Access Amid Government Interest

Anthropic’s newly released Mythos AI model has been accessed by a modest group of unauthorized users, according to Bloomberg News reporting cited by multiple outlets on April 21, 2026. The model, described by the company as powerful enough to enable dangerous cyberattacks, remains in a limited release phase despite growing interest from government agencies.

Bloomberg reported that the unauthorized access was identified through documentation and confirmed by a person familiar with the matter. Anthropic has not disclosed how the breach occurred or whether any data was compromised. The company maintains strict controls over Mythos, which it says is designed with advanced safety protocols due to its potential dual-use capabilities in both civilian and military applications.

Despite the security incident, federal interest in the model continues. A separate Forbes report from April 16, 2026, indicated that the White House is planning to provide federal agencies with a version of Mythos, citing an internal memo from Gregory Barbaccia of the Office of Management and Budget. The memo outlines efforts to establish protections that would allow government apply, though it does not explicitly confirm access details or usage parameters.

Anthropic co-founder Jack Clark has previously stated that the company is engaged in discussions with the U.S. Government about Mythos, emphasizing that officials need to understand the technology’s implications for national security and cybersecurity. These talks come amid ongoing friction with the Department of Defense, which designated Anthropic a supply-chain risk after the company declined to provide unrestricted access to its AI models.

Adding to the complexity, the National Security Agency (NSA) is reportedly using an advanced variant of Anthropic’s AI technology—referred to as Claude Mythos—despite earlier Pentagon warnings about supply-chain risks. This development highlights a divergence in how different U.S. Government entities are engaging with Anthropic’s most capable models.

As of April 21, 2026, Anthropic has not issued a public statement addressing the unauthorized access incident or detailing steps taken to reinforce Mythos’s security controls. The company continues to limit distribution of the model to a select group of trusted technology partners while navigating heightened scrutiny from both government and industry stakeholders.