Deepfakes became significantly more difficult to spot in 2025, and a University of Buffalo researcher predicts the trend will continue in 2026.