Show HN: Video Analysis with Homomorphic Encryption (github.com)

Belfort Labs has introduced BlindSight, a groundbreaking video analysis system that utilizes fully homomorphic encryption (FHE) to detect the presence of a person in a webcam stream without ever exposing the video frames. This solution enables the client to encrypt video frames locally before sending them to the Belfort eML server for inference, ensuring that sensitive visual data is never exposed, as the server processes the encrypted frames and returns an encrypted verdict back to the client. This innovation is significant for the AI/ML community as it represents a major step toward enhancing privacy in machine learning applications, enabling safer deployment of AI systems in sensitive environments. The underlying architecture leverages a lightweight convolutional network called PersonNet, which operates seamlessly under FHE. This means that the entire process—capturing, encrypting, processing, and decrypting—occurs without revealing the images, thus potentially transforming applications in security, healthcare, and personal privacy where data confidentiality is paramount. With optimized performance that allows inference times to potentially drop below one second, BlindSight demonstrates both the feasibility and utility of secure AI for real-world applications.