Stream Engine Develops AI Deepfake Detection for Air Force

• •Stream Engine delivered a deepfake detection analytic to the Air Force.
• •The analytic uses a two-stage detection pipeline based on semantic analysis.
• •The project faced operational challenges but achieved test readiness in June 2026.

Stream Engine has delivered a semantic-based deepfake detection analytic to the Air Force Research Laboratory (AFRL) under a Cooperative Research and Development Agreement. This analytic, designed to identify machine-generated text, has been integrated into AFRL's operational Test and Evaluation testbed, which became ready for formal testing in June 2026. The project aimed to enhance the detection of deepfakes by analyzing grammatical, syntactic, and semantic features of text. Stream Engine utilized its expertise in large language models to create a two-stage detection pipeline. The collaboration faced challenges, including a federal government shutdown and facility power outages, but successfully reached test-ready status. No federal funds were exchanged, with Stream Engine's contributions estimated at over $1 million.