GSA Blog

Making access to AI a priority

How FedRAMP used public engagement to build a framework for safe and easy AI adoption

Keeping up with emerging tech is no easy feat, especially with artificial intelligence increasing in our daily lives. From automated assistants to self-driving cars, AI has made significant impacts on how we live and in our daily routines.

For government agencies, the question about AI has evolved into: “How can we adopt AI responsibly to better serve our nation?” GSA has been at the forefront of providing access to cutting-edge technology for decades. Streamlining access to new and emerging tech to make government work better is a cornerstone of the Federal Risk and Management Program (FedRAMP). 

A framework for governmentwide adoption

Today, GSA released the final Emerging Technology (ET) Prioritization Framework for how FedRAMP will prioritize emerging tech, beginning with generative AI. The final framework was developed in response to the Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence and incorporates feedback received on the draft framework we issued earlier this year. 

In less than two months, we received over 200 comments on the draft framework. We greatly appreciate everyone who took the time to contribute their perspective to strengthen the final product.  

Today, FedRAMP opened the first application period and qualifying cloud service providers (CSPs) can request a priority review of their in-demand offering now through August 31, 2024. Prioritized cloud services will complete the normal FedRAMP authorization process, but will get an accelerated start on that process so federal agencies can more quickly get access to emerging technology through the FedRAMP Marketplace. Initially, we expect up to twelve AI-based cloud services to be prioritized using this framework. 

Listening to our customers 

Customer feedback gave the FedRAMP team  important information to guide how we handle the first prioritized technology, generative AI. 

Here is what changed in the overall framework approach:

• Streamlined content - The final framework offers a simplified web-based experience.
• Improved flexibility -  The number of prioritized services can be increased to respond to agency demand.
• Demand forecasting - An adjusted prioritization demand formula focuses on CSPs with current demand, not just future demand.

Here is what changed in FedRAMP’s approach to our first emerging technology, generative AI:

• Moving from benchmarks to AI model cards - The AI services will no longer be evaluated on pre-defined benchmarks, but industry-standard AI model cards that describe the service offering’s intended use cases and limitations. 
• Supporting APIs - Now includes prioritization slots for offerings primarily focused on incorporating generative AI into new or existing systems through Application Programming Interfaces (APIs).
• Training update - The use of government data for general AI training by AI services prioritized using this framework must be opt-in, ensuring AI meets the security and privacy needs across agency use cases.

Our prioritization framework will result in additional generative AI offerings being added to the FedRAMP Marketplace, allowing rapid adoption by agencies to meet both their enterprise and public mission. The framework will prioritize offerings that include: 

• Chat interfaces.
• Code-generation and debugging tools.
• Prompt-based image generators.
• General-purpose API offerings of the above three capabilities.

Moving forward, FedRAMP will continue to identify additional emerging technologies for prioritization. For more information about FedRAMP and technical details about the final framework, please read the Release of the Emerging Technology Prioritization Framework blog post on fedramp.gov.