Pentagon projects with AI sometimes ignore its own responsible AI guidelines

A congressional watchdog report published in October reveals that some weapons and projects developed by the Pentagon with artificial intelligence (AI) capabilities do not always adhere closely to the agency's own responsible AI guidelines, according to krvs.org and wdiy.org.

The Government Accountability Office (GAO) report surveyed 10 officials involved in building AI-enabled weapons and tools, including those that automatically identify battlefield targets, process surveillance, and assist tactical decision-making. Among these officials, six acknowledged identifying risks when using their projects in operational environments, as reported by krvs.org.

In response to such risks, the Pentagon issued a memo in 2021 outlining six basic tenets for responsible AI use in weapons development. These guidelines emphasize agency-wide oversight, verification, real-time monitoring, and cooperation with allies to promote global norms based on shared values, according to krvs.org.

Despite these guidelines, the Pentagon's push for rapid autonomous innovation continues. Defense Secretary Pete Hegseth has promoted making the department an 'AI-first' warfighting force and recently announced a command focused on autonomous weapon development, including drones, as noted by krvs.org.

The GAO report provides rare insight into the development of over 300 Pentagon projects involving AI, highlighting the challenges in balancing innovation with responsible use of AI in military applications.

What we're watching: How the Pentagon will address the identified risks and adherence gaps as it advances AI-enabled weapons development.