Innovate UK CyberASAP Pathfinder 2025: SARAD: Secure AI For Robust Anomaly Detection

Project Description

Anomaly detection is essential across industries for fraud detection in finance, identifying equipment malfunctions in manufacturing, and detecting abnormal health conditions in healthcare. AI-based anomaly detection is state-of-the-art, but to fully realize its potential, it must be secure and responsible. AI systems (including AI anomaly detection) face new security risks, e.g., poisoning (corrupting training data, thus degrading AI's ability to learn correctly) and evasion (tricking AI into failing to recognize anomalies) attacks, alongside traditional threats. In March 2024, a survey by HiddenLayer.com found that 77% of IT security and data science leaders reported AI breaches last year. In May 2024, the UK government introduced two new codes of practice to enhance AI cybersecurity, and supporting the growth of the UK's cyber sector. In line with the government policies, our project’s idea is to develop Secure-AI Anomaly Detector. This cloud-based solution leverages a secure, explainable AI model to detect and mitigate anomalies in time-series data. It can efficiently process large datasets, identify anomalies at any scale, and take proactive steps to mitigate them. By combining advanced AI techniques (robust AI security measures and transparency), this solution delivers both accurate predictions and strong protection. Moreover, its built-in explainability features clarify how decisions are made, fostering greater trust and confidence in AI-driven systems.