Some Pentagon projects that use artificial intelligence do not closely follow the department's own responsible AI guidelines, according to a Government Accountability Office report published on 14 October, NPR reports.
The watchdog surveyed officials working on ten AI weapons and tools, including systems that identify battlefield targets and process surveillance, out of more than 300 such projects. Six said their teams had identified risks in operational use, four said their requirements were not aligned with the tenets, and three did not know whether their projects would be monitored after deployment.
The six tenets, set out in a 2021 memo, stress oversight, verification and real-time monitoring, but are guidance rather than binding instructions. The report found the office responsible for training on them had filled only three of ten authorised posts. Defence Secretary Pete Hegseth has pushed for rapid adoption of autonomous technology.
Source: NPR News
