The National Institute of Standards and Technology (NIST) is taking another step toward operationalizing its artificial intelligence (AI) safety agenda and is seeking companies that can host and run open-weight AI models for federal evaluation efforts.
The “sources sought” notice published to SAM.gov on Tuesday by NIST is to gauge private-sector capability to support the Center for AI Standards and Innovation (CAISI) and its work on evaluating advanced AI systems for national security risks.
CAISI was tasked with those evaluations under the Trump administration’s AI Action Plan, which placed on emphasis on examining how the most powerful AI systems may pose new national security risks in the near future. Specifically, CAISI was tasked to evaluate frontier AI, in conjunction with other agencies with expertise in cyber risks.
“In order to deliver on this tasking, CAISI requires access to open-weight AI model hosting and inference services,” NIST said in the notice.
NIST said that it needs vendors that can quickly stand up newly released open-weight models, provide accurate and validated inference services, scale usage rapidly, and maintain high reliability.
The agency also signaled a strong interest in data safeguards, specifying that inference data should not be retained.
NIST said responses to the notice will help inform its acquisition strategy, including whether a future competition could be structured with a small-business set-aside. Responses are due by March 10.
The agency said it anticipates issuing a formal request for quotation in the second quarter of fiscal year 2026, with a potential award by the third quarter.