Enhancing the Security, Privacy and Robustness of AI Models and Systems (SecureAI)

Expected Outcome: Proposals are expected to contribute to one or more of the following: Robust AI models and systems capable of resisting different classes of adversarial manipulation; Innovative defence mechanisms for AI models and systems against new attack families; Methodologies for detecting and mitigating attacks, such as data poisoning, backdoor exploitation and misclassification; AI systems leveraging privacy-enhancing technologies that maintain data confidentiality and regulatory compliance, enabling trusted in-house AI deployments (e.g., for governments and enterprises). Scope: The increasing reliance on AI in cybersecurity, critical infrastructure, and decision-making processes raises concerns about the security and robustness of AI systems. As AI systems become more prevalent, they are increasingly targeted by adversarial attacks that manipulate inputs, compromise training data, or introduce hidden vulnerabilities. This topic aims to strengthen the resilience of AI systems and algorithms against various threats and attacks, such as enhancing their resilience against adversarial attacks, backdoor injections, and data poisoning. Proposals should develop real-time anomaly detection, mitigation techniques to defend against adversarial attacks and robust federated learning techniques, in synergies with leading efforts on AI transparency, and in compliance with the AI Act. The topic is expected to: Develop robust AI models resistant to adversarial attacks. Exploring…