Multiple models concurrently detect AI generation and manipulation in audio, video, images, and text. Results are probabilistic, rather than deterministic, which means the models don't require watermarks or prior authentication to test for authenticity.