Researchers from the University of California, Riverside, YouTube, and Google DeepMind have developed an artificial intelligence system called SAGA (Source Attribution of Generative AI Videos) that can identify whether a video was generated by AI and determine which AI model created it, according to a newly published research paper. The system attributes videos to a specific generator, development team, model version, or generation method across five levels of identification: distinguishing real from AI-generated videos, identifying whether a video was created from text or an image prompt, recognizing the underlying model version, identifying the development team, and pinpointing the exact AI generator. SAGA learns from very limited labeled data, achieving results comparable to fully supervised methods while using only 0.5% of the source-labeled training data for each AI model. The researchers introduced Temporal Attention Signatures, visual patterns revealing subtle differences in how AI video generators create motion, which may help identify videos from previously unseen generators. The team tested SAGA on public datasets containing videos generated by 19 different AI models, and the framework consistently outperformed existing video detection methods, remaining effective on videos from generators not encountered during training. Researchers say SAGA could strengthen digital forensics and improve transparency as AI-generated video becomes more widespread.