AI Aware Detector Accuracy: Results Tables for Text, Image/Video and Audio Detectors (August to November 2026)

Published: 12 November 2026| Version 1 | DOI: 10.17632/bwwcjrb398.1
Contributor:
Stephen Harmston

Description

Results behind AI Aware's published accuracy figures, covering the best result from each detector between 14 August and 7 November 2026. Text: 98.6% accuracy (AUROC 0.998) on 10,334 held-out texts from 30 AI models across 14 types of writing. Images and video: 99.4% accuracy (AUROC 0.9996) on 43,106 held-out images from 28 generators, with 0.39% of real photos wrongly flagged. Audio: 97.6% accuracy (AUROC 0.995) on the BRSpeech-DF validation set. The CSV files give summary metrics and breakdowns by type of writing, text length, AI model, image generator and source of real photos. The PDF explains the method. See 00-README.txt for a list of files. Full results and method: https://aiaware.io/ai-detector-accuracy-technical-results Comparison with other detectors: https://aiaware.io/most-accurate-ai-detector

Files

Categories

Artificial Intelligence, Machine Learning, Deepfake

Licence