Horizon Cyber: Robust & Private AI Systems
Funds EU teams hardening AI systems against adversarial attacks and building privacy-preserving AI for sensitive government and enterprise use.
AI is now woven into cybersecurity, critical infrastructure and decision-making — which makes AI itself a target. Attackers manipulate inputs, poison the data a model trains on, or hide backdoors that only trigger later. This topic funds work that makes AI systems hold up against that.
A winning project can go after robustness, detection, or privacy — the call asks for one or more, not all three.…
Building AI models and systems that hold up against adversarial manipulation — attacks that poison training data, hide backdoors, or trick a model into misclassifying — and building privacy-preserving AI that governments and enterprises can run in-house without exposing the underlying data. A proposal can target one or more of: robust models, new defences against emerging attack families, detection of poisoned or backdoored training data, or Private AI using techniques like federated learning, secure aggregation, computation on encrypted data, and secure inference.
Sign up free to see the funding breakdown
Sign up free to see the industries in scope
Sign up free to see the full eligibility
Sign up free to see how to apply
Sign up free to see what you submit
Sign up free to see how they score you
Sign up free to see the timeline
Sign up free to see where teams trip up