Federal agencies are exploring access to Anthropic’s powerful recent AI model, Mythos, despite ongoing tensions between the company and the Pentagon over national security concerns. Discussions between the White House and Anthropic have intensified as civilian departments seek to leverage the model’s advanced cybersecurity capabilities for defensive purposes, even as legal disputes with the Department of Defense continue.
The conversations reflect a growing divide within the U.S. Government over how to balance caution with innovation when deploying frontier AI systems. Even as the Pentagon has labeled Anthropic a supply chain risk and restricted its involvement in military contracts, other agencies see value in Mythos for protecting critical infrastructure like the energy grid and financial systems.
According to Bloomberg News, the White House Office of Management and Budget has begun evaluating how to safely introduce a version of Mythos to federal agencies, emphasizing the necessitate for guardrails before any broader release. Gregory Barbaccia, federal chief information officer at the OMB, stated in an internal email that officials are working with model providers, industry partners, and the intelligence community to establish appropriate safeguards.
Anthropic’s Mythos model, announced on April 7, 2026, is being rolled out under a controlled initiative called “Project Glasswing,” which permits select organizations to test the unreleased Claude Mythos Preview for defensive cybersecurity purposes. Reports indicate the model has demonstrated an ability to identify thousands of vulnerabilities in operating systems, web browsers, and other software, raising both interest and concern across government circles.
White House Advances Talks Amid Pentagon Dispute
Despite the Pentagon’s designation of Anthropic as a supply chain risk—a label stemming from an ongoing legal battle—the White House is pursuing separate discussions about potential Mythos access for non-defense agencies. Officials familiar with the matter told Axios that while progress has stalled with the Department of Defense, talks with the White House are advancing, with some agencies possibly gaining access in the coming weeks.
The Pentagon’s stance centers on Anthropic’s refusal to permit its models to be used for mass surveillance or fully autonomous weapons development, which the Defense Department views as overly restrictive. Anthropic maintains that these boundaries are necessary to prevent misuse, while the Pentagon argues it needs flexibility to employ AI for “all lawful purposes.”
One administration official noted that intelligence agencies broadly support Anthropic’s tools precisely as they avoid applications tied to lethal autonomy, contrasting with the Pentagon’s position. Another official acknowledged the government’s responsibility to evaluate cutting-edge models but accused Anthropic of using “fear tactics” by highlighting Mythos’s potential to enhance offensive hacking capabilities if misused.
Civilian Agencies See Strategic Value in Mythos
Departments such as Energy and Treasury, which oversee critical sectors including the electric grid and financial infrastructure, have expressed particular interest in Mythos. Access to the model could help these agencies identify weaknesses in software used by private companies and local governments, enabling them to issue timely advisories and strengthen resilience against cyber threats.
This focus on defensive cybersecurity aligns with the stated purpose of Project Glasswing, which aims to let trusted organizations stress-test systems before adversaries can exploit flaws. Anthropic has emphasized that Mythos is not being released publicly due to its potential to accelerate vulnerability discovery—and, conversely, exploitation—if placed in the wrong hands.
The model’s capacity to generate sophisticated code has drawn attention from experts who warn that its dual-use nature requires careful oversight. While its ability to detect flaws could improve national cyber defenses, the same skill could be repurposed to develop advanced hacking tools, a concern cited by officials wary of accelerating offensive cyber capabilities.
Legal and Policy Tensions Shape Access Debate
The friction between Anthropic and the Pentagon remains unresolved, with the company challenging the Defense Department’s supply chain risk designation in court. Until a legal resolution, Anthropic is barred from securing new Pentagon contracts but continues to operate with other federal entities.
This split has created an unusual dynamic where certain branches of government are moving forward with AI adoption while others remain locked in dispute. Officials familiar with the internal debate described a sentiment among some Trump administration figures who view Anthropic leadership as ideologically driven—referred to privately as “woke doomsters”—yet still acknowledge the technical superiority of its models.
One Defense official, speaking during the height of the feud, reportedly told Axios that the only reason negotiations had not collapsed entirely was Anthropic’s undeniable capability: “these guys are that good.” The comment underscores a broader tension in U.S. Policy circles between distrust of certain tech firms and recognition of their strategic value in emerging technologies like AI.
As of April 16, 2026, no formal decision has been made regarding Mythos access for federal agencies beyond the ongoing evaluations led by the OMB. The administration continues to stress that any deployment will depend on verified safeguards, though specific timelines or conditions have not been disclosed.
For updates on federal AI policy and agency evaluations of frontier models, readers can refer to official releases from the White House Office of Management and Budget and public statements from the Department of Defense regarding AI ethics and acquisition guidelines.