Pentagon Commences Bounty Program to Identify AI Bias

The Department of Defense’s Chief Digital and Artificial Intelligence Office has kicked off the first of two bounty programs to identify biases in artificial intelligence systems.

Led by the CDAO Responsible AI division, the AI Bias Bounty exercises are expected to produce techniques to audit and red-team AI models algorithmically, the DOD said.

Matthew Johnson, the RAI division’s acting chief and a speaker at the Potomac Officers Club’s upcoming 5th Annual Artificial Intelligence Summit, affirmed his team’s commitment to eliminating bias in the DOD’s AI-enabled systems “and the contexts in which they run.”

Johnson’s office is managing the AI Bias Bounty program in partnership with ConductorAI-Bugcrowd and BiasBounty .AI and with advisory support from the CDAO Defense Digital Services Directorate.

Ongoing until Feb. 27, the first exercise focuses on identifying unknown risks in large language models, including open-source chatbots. It is open to public participation, with monetary prizes to be given out based on scoring by ConductorAI-Bugcrowd.