Federal agencies could get modified Anthropic Mythos access under new White House plan
The US government is poised to introduce a tailored version of Anthropic's cutting-edge Mythos AI model to key federal agencies, as it navigates the complexities of leveraging a system that has sparked fresh cybersecurity concerns. In a recent memo, Gregory Barbaccia, the federal chief information officer, apprised senior technology and cybersecurity officials of an impending announcement regarding potential access to the model, slated for release in the coming weeks.
Although the details of the rollout remain unclear, the memo hints at the development of robust safeguards to mitigate potential risks associated with the model's deployment. The communiqué, which was circulated to top officials from various departments, including Defense, Treasury, and Homeland Security, underscores the government's enthusiasm for harnessing the potential of frontier AI systems to bolster federal cyber defense capabilities.
Notably, the Treasury Department has been seeking to utilize Mythos to identify software vulnerabilities, as reported by Bloomberg on April 14. This development is particularly significant, given that Mythos is not a commercially available model, but rather a limited research preview, introduced by Anthropic on April 7 as part of its Project Glasswing initiative. The company has emphasized that Mythos boasts unparalleled strength in cybersecurity, which is precisely why it is being kept under strict control.
As part of Project Glasswing, Anthropic is collaborating with a select group of partners to leverage Mythos to enhance critical software security and prepare institutions for the potential risks and benefits of frontier AI models. The company has been engaged in ongoing discussions with US government officials regarding Mythos and its cyber capabilities, with a focus on protecting critical infrastructure, which is deemed a national security priority.
The restricted release of Mythos follows widespread concern within both government and financial circles, prompted by warnings from Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell regarding the potential risks associated with the model. In a recent development, Anthropic has released Claude Opus 4.7, a model that is less capable than Mythos but features built-in safeguards to detect and block high-risk cybersecurity requests. The company intends to draw lessons from the real-world deployment of Opus 4.7 to inform the potential broader release of Mythos-class models.
This latest development is particularly noteworthy, given that the US government had previously directed federal agencies to cease using Anthropic's technology, as announced by Trump on February 27. The new discussions surrounding Mythos suggest that officials may be seeking to reintroduce Anthropic's technology, albeit in a controlled manner, specifically for cybersecurity applications.