“It’s like the Wild, Wild West when it comes to AI right now,” Linchpin product lead Bharat Patel told Breaking Defense. “We don’t want to be the government and be like, ‘Here’re our standards, you must comply.’”
Instead of demanding an exhaustive “AI Bill of Materials.” the Army will only ask contractors for a “baseball card” of key stats on their AI — while building up its in-house capacity to check for bad code or “poisoned” data.
Balancing security and access to data are critical to providing users with trustworthy, reliable data for real-time decisions.
“Any commercial LLM that is out there, that is learning from the internet, is poisoned today,” Jennifer Swanson said, “but our main concern [is] those algorithms that are going to be informing battlefield decisions.”
The service is also currently developing a risk management framework for Project Linchpin, the Army’s first program of record to help build out a trusted artificial intelligence/machine learning pipeline, according to Jen Swanson, deputy assistant secretary of the Army for data, engineering and software.
“The goal is to create a complete and efficient AI and ML development and delivery operational pipeline (AI/MLOps) with supporting services for sensor programs within PEO IEW&S while managing cost, schedule, risk and performance," according to a request for information.
The service “will use Project Linchpin to develop and deploy trusted AI & ML capabilities to intelligence, cyber, and electronic warfare sensor systems,” Col. Chris Anderson, project manager for intelligence systems and analytics, said.
“Tactical operating system” for warfighters has evolved to provide more situational awareness of the battlefield.
“I think across our entire portfolio, there are opportunities for us to simplify and deliver capabilities that allow for our units to have a more flexible network infrastructure,” Mark Kitz said.
"The ambition of [Project] Linchpin makes me nervous," said Mark Kitz, program executive officer for IEW&S. "We can't answer all of our AI/ML ops questions with the initial iteration of Linchpin."