BitMind has announced that its deepfake detection system, BitMind Forensics, ranks among the world’s best, according to a newly published paper. The system matched the performance of the top commercial image detector and surpassed the best commercial video detector while excelling in various major benchmarks. Utilizing Bittensor’s SN34 subnet, BitMind engages in continuous adversarial training, leading to the development of increasingly effective detectors as generative capabilities evolve. The detection system is available through multiple platforms, including a web app, mobile app, Chrome extension, and API, catering to both individual users and enterprise clients.

SN34: SN34 is a Bittensor subnet that hosts an open adversarial competition for deepfake detection. Generative miners create synthetic images and videos to fool detectors, while discriminative miners submit detection models that validators test and reward based on performance. The subnet’s outputs directly power the iterative improvements in BitMind Forensics.
BitMind: BitMind develops deepfake detection technology powered by the Bittensor decentralized AI network. The project created SN34 as an open adversarial subnet where generative miners produce synthetic media to challenge detectors submitted by discriminative miners. This continuous competition feeds into BitMind Forensics, the production system that delivered state-of-the-art results on recent benchmarks.
Bittensor: Bittensor is a decentralized network that supports specialized AI subnets for collaborative machine learning tasks. BitMind utilizes it to run SN34, an adversarial competition that generates harder training data and stronger detectors for deepfake detection through validator rewards.
BitMind Forensics: BitMind Forensics is the production deepfake detection system built from outputs of the SN34 subnet on Bittensor. It integrates the strongest detectors and latest synthetic data from ongoing miner competitions, enabling consistent performance across multiple public datasets without per-benchmark retraining. The system matched the top commercial image detector and outperformed the leading video detector in independent tests.

Adversarial Training: BitMind’s SN34 subnet runs continuous competitions between generative and discriminative miners to produce harder training data and stronger detectors as generators evolve.
Product Availability: BitMind Forensics is accessible via web app, mobile app, Chrome extension, and API for both individual users and enterprise integration.
Benchmark Performance: BitMind Forensics matched the best commercial image detector and beat the best commercial video detector while outperforming top models on major benchmarks.