DoD requests $30.3 million for Polygraph+ AI scoring and standoff sensing

The request funds development of automated polygraph scoring and contactless sensors that capture heart rate, respiration and other signals at distance. Prior polygraph programs relied on subjective examiner interpretation with documented error rates above 20 percent in controlled studies. The new program combines these legacy signals with machine learning models trained on larger datasets while adding remote sensing to reduce setup time.

Historical data from National Academy of Sciences reviews show standard polygraphs produce high false-positive rates when scaled beyond laboratory conditions. AI versions risk amplifying those errors through overfitting to narrow training populations and lack of adversarial testing. Standoff sensing introduces additional variables such as clothing, lighting and subject movement that have not been validated at operational volumes.

Operationally the system would feed into security clearance and counterintelligence workflows already handling thousands of examinations annually. Integration with existing databases creates persistent risk scores that follow individuals across agencies without clear appeal mechanisms. Past attempts at automated deception detection, including layered voice analysis pilots, were scaled back after independent audits found no reliable improvement over chance.

Field deployment thresholds remain unspecified in the request, leaving open whether accuracy must exceed current baselines before operational use. Ethical review processes for defense AI programs have historically lagged hardware milestones by multiple years.