The government-backed nonprofit organization Mitre opened a lab Monday to test AI systems used by federal agencies, aiming to discover and fix security flaws and other risks.

The lab will assess the systems for everything from data leaks to explainability — the ability to see why AI technology is making certain decisions — according to Miles Thompson, a robotics engineer who will head the lab. The facility is based at the McLean, Va., headquarters of Mitre, which oversees research on national security, aviation, health and cybersecurity, among other topics. It can accommodate 50 people in person and 4,000 via remote connections.

Experts have warned that AI systems, which are often likened to a black box, are being adopted without fully understanding the myriad ways they can be tricked or go wrong.

In a sign of how quickly the U.S. government is embracing AI, Thompson said he and co-workers have been frequently asked to do “one-off” assessments of the technology. So the organization decided to create an entire lab to handle the process.

The new facility, called the AI Assurance and Discovery Lab, will attempt to learn the risks of AI systems by hacking them and testing them for bias. Sen Mark Warner, chairman of the Select Committee on Intelligence, was among senior lawmakers who visited the lab for its opening on Monday. Warner, a Democrat from Virginia, described it as an effort to extract maximum value from AI while mitigating some of its risks.

“We need to have an all-hands-on-deck approach to studying and unleashing the potential of AI, and I look forward to seeing the discoveries and progress the lab will be able to make in this critical field,” he said in a statement.

©2024 Bloomberg L.P.