Brandon Fletcher
A recent report by Politico reveals that the Pentagon aims to integrate and potentially weaponize advanced AI models, particularly those with hacking capabilities, such as the Claude Mythos Preview.
This information stems from a leaked email and insights from two anonymous sources, according to Politico.
Officially, the Pentagon is not permitted to use Claude Mythos Preview in an open capacity. Anthropic’s forthcoming model is considered too powerful for public release, making it particularly appealing to military interests. the Pentagon has recently deemed Anthropic a supply chain risk following a significant public dispute, limiting its collaboration with the company. This designation suggests that not only is the Pentagon barred from engaging with Anthropic, but its contractors face similar restrictions.
Joshua Rudd, the leader of Cyber Command and director of the NSA, announced this new AI initiative earlier this month. Both agencies are reportedly collaborating on this task force, which has goals reminiscent of Anthropic’s Project Glasswing. This project enables select companies and organizations to utilize Claude Mythos Preview to identify and repair vulnerabilities within their systems.
According to Politico, the task force is set to “examine how the Pentagon can securely implement leading AI models across all mission areas” and evaluate “the potential use of AI models developed by Silicon Valley tech firms on ‘high-side’ systems that protect some of the intelligence community’s most sensitive information.”
The report suggests that this Pentagon task force may also explore dimensions beyond those currently addressed by Glasswing, specifically the potential weaponization of AI models equipped with hacking capabilities. Former Cyber Command deputy commander, Charles Moore, remarked to Politico, “AI tools are becoming crucial for identifying threats, prioritizing vulnerabilities, speeding up decision-making, and conducting both defensive and offensive cyber operations more swiftly than our adversaries.”
Politico also mentions that Anthropic has previously indicated its willingness to allow Mythos to be utilized in offensive operations.
Despite the existing tensions with Anthropic, the Trump Administration has shown some involvement with Project Glasswing. A Wall Street Journal article from late last month reported that the White House is “engaged in the deployment of Mythos due to the national-security risks associated with the model,” and had exercised veto power over Anthropic’s proposal to expand Project Glasswing by adding 70 new companies and organizations.
According to Politico, the Pentagon is “likely to investigate” the potential applications of Mythos.
Daniel Mercer
