NGA wants to assess AI/ML supply chain risks for Project Maven

WASHINGTON — The National Geospatial Intelligence Agency (NGA) wants to build out a platform to assess supply chain risks posed to artificial intelligence and machine learning technologies for Project Maven, an effort that aims to speed up the use of AI across the military. 

According to a new NGA solicitation, the platform will need to map all vendor links to each other and discover any “problematic relationships within vendor networks.” 

“Maven needs [a] quick-turn open-source, unclassified reporting system that allows NGA to share information of concern with the vendor and the Government without possible inadvertent disclosure of vendor proprietary information,” the April 24 solicitation says. “The platform will also be available as a repository for mission reporting and information sharing about high and critical threats within agency’s supply chain.”

Project Maven is slated to become a formal program of record in fiscal 2024. According to the solicitation, NGA needs to assess AI supply chain risks for Project Maven “because there is often limited visibility into the prime vendors and the sub-tier suppliers that provide critical components and personnel.” 

“Further, AI/ML supply chain illumination enables the identification of opportunities, vulnerabilities, and systemic dependencies that are crucial to assessing AI/ML vendor health and making sound, secure prototype and acquisition decisions,” the solicitation says. 

NGA has listed specific tasks the supply chain risk management platform should be able to do, including: rapid network mapping of vendor relationships, organizing a data-labeling effort, documenting unclassified information to share with vendors, and integrating “algorithmic-based technology with Programs of Record, a rapid prototyping effort that adapts, prototypes, and integrates commercial AI technology into DoD platforms with active tactical users.”

Project Maven, which was established in 2017, was shifted to the NGA last year from the Office of the Under Secretary of Defense for Intelligence and Security. As the Pentagon described it at its founding, the project “focuses on computer vision — an aspect of machine learning and deep learning — that autonomously extracts objects of interest from moving or still imagery. […] Biologically inspired neural networks are used in this process, and deep learning is defined as applying such neural networks to learning tasks.”

While the NGA has operational control of the program’s GEOINT services, the Pentagon’s Chief Digital and AI Office was also designated to take some parts of Project Maven in President Joe Biden’s fiscal 2023 budget request to Congress.

Last April, NGA’s then-Director Vice Adm. Robert Sharp said a key focus area for the agency would be expanding Project Maven to combine the “computer vision” with human expert analysis.

“Starting next fiscal year, will have complementary computer vision efforts that will deliver automated GEOINT detections to both intelligence analysts and warfighters,” Sharp said then. “In our key role as a Combat Support Agency, we’ll provide the subject matter experts — humans who can train the machine evaluate it and make sense of the output. We’ll bring together those disparate sometimes siloed communities of machine learning experts, data scientists, GIS experts and imagery analysts to improve AI model performance, develop standards and lead interoperability efforts for the GEOINT community.”