CAISI has been ordered to cease public model evaluations, a decision influenced by the implementation of a new AI Executive Order. This policy shift is part of a broader government initiative responding to heightened concerns over national security as advanced AI models, such as Mythos, are released. Critics argue that restricting access to CAISI’s evaluations diminishes scrutiny and transparency in AI science, suggesting that it is possible to balance national security with the need for openness in the field.

CAISI: CAISI is the Center for AI Standards and Innovation, a unit housed at the U.S. National Institute of Standards and Technology (NIST) that conducts evaluations of frontier AI models, with emphasis on cybersecurity, agentic capabilities, and national security implications. It collaborates with industry and other agencies on testing methodologies and has previously published public assessments of models. The center has been directed to cease public model evaluations amid implementation of a new AI Executive Order, a shift reportedly influenced by the release of Anthropic’s Claude Mythos.
Janet Egan: Janet Egan is a policy expert at the Center for a New American Security (CNAS) specializing in AI, compute, and national security issues. She publicly commented on the CAISI directive, arguing that reducing public visibility into model evaluations undermines broader security goals and that openness and national security objectives can coexist.

`json
{
“Policy Shift”: “The U.S. government has initiated changes in the sharing of AI evaluation outcomes due to a new Executive Order.”,
“National Security Focus”: “The release of advanced AI models, such as Mythos, has increased government scrutiny on AI, emphasizing assessments of capabilities and risks.”
}
`