The U.S. government is preparing to let some federal agencies use a modified iteration of Anthropic's frontier AI model, Mythos, under a tightly controlled program intended for cybersecurity defense, according to reporting based on internal communications.
Unveiled on April 7 as part of Anthropic's Project Glasswing, the unreleased Claude Mythos Preview model is being made available to select organizations for defensive purposes. The model has reportedly identified "thousands" of significant vulnerabilities across operating systems, web browsers and other widely used software, demonstrating an ability to generate high-level code and to surface potential exploits.
Gregory Barbaccia, federal chief information officer at the White House Office of Management and Budget (OMB), sent an email to Cabinet department officials saying the OMB is establishing protections so agencies can begin using Mythos. The message, whose subject line read "Mythos Model Access," stated: "We’re working closely with model providers, other industry partners, and the intelligence community to ensure the appropriate guardrails and safeguards are in place before potentially releasing a modified version of the model to agencies."
The email does not confirm that agencies will receive access, nor does it include a schedule for when access might be granted or details on operational use by agencies. Those items remain unsettled in the communications reviewed.
Observers quoted in the reporting have noted that Mythos' advanced coding capabilities give it an unusually potent capacity to identify software weaknesses and to propose methods of exploitation. That potency is the principal reason the model is being introduced under a controlled framework - to use its defensive detection strengths while attempting to limit the risk that its outputs could be repurposed for offensive attacks.
Separately, Anthropic co-founder Jack Clark has said the company has been in discussion with the current administration. Those conversations have continued even after another federal client, the Department of Defense, halted business with Anthropic following a contract dispute.
The White House and Anthropic did not immediately provide comment in response to requests for comment. Key uncertainties remain around the timing of any agency access, the precise safeguards that will be applied, and how agencies would operationalize Mythos within existing cybersecurity programs.
Implications
- Government cybersecurity teams could gain a potent vulnerability-detection tool if access is approved and safeguards are effective.
- The same technical strengths that make Mythos useful defensively raise concerns about increased cybersecurity risk if outputs are misused.
- Federal contracting relationships with AI vendors remain subject to dispute and review, as illustrated by the pause in business with the Department of Defense.