Research direction Active
Secure and Auditable AI Agents
Finding, measuring and containing the security risks of LLM agents and their skill and tool ecosystems, so that agents acting on behalf of people and organisations can be trusted and audited.
Research directions and demonstrators6
Research direction Proposed
Privacy-Preserving Cyber-Threat Intelligence Sharing
Letting organisations pool indicators, incidents and attack analytics without exposing the sensitive data behind them, using secure computation, differential privacy and cryptographic access control.
Research direction Proposed
Secure Collaborative Analytics
Practical secure multi-party computation, homomorphic encryption and trusted-execution designs that let several organisations compute joint insights on health, financial and security data without revealing their inputs.
Research direction Active
Verifiable Provenance for Software, Data and AI Supply Chains
Cryptographic provenance, watermarking and continuous analysis that show where software, data and AI models came from, whether they were tampered with, and whether they can be trusted.
Research direction Active
Continuous Assurance and Red-Team Pipelines
Reusable benchmarks, automated penetration testing, fuzzing and adversarial evaluation that measure whether software, networks and AI systems stay secure as they change.
Research direction Active
Resilient Security for Critical Infrastructure and Energy Systems
Security architectures for distributed energy assets and other critical infrastructure, where AI-enabled control, federated data and physical consequences meet.
Funded projects2
Funded project Active
Secure Federated Battery Management System for Distributed Energy Storage Assets
A privacy-preserving, attack-resilient federated learning platform for battery health and safety across fleets of distributed energy storage assets.
Funded project Active
Trustworthy, Energy-Aware Computing Load Balancing for a Distributed Residential GPU Network
Secure scheduling and telemetry integrity for a network of residential GPUs that trade compute against local energy availability.
Software and artefacts2
Software Active
PentestGPT
An open-source, LLM-powered penetration-testing assistant that plans, reasons and keeps track of an engagement, evaluated at USENIX Security 2024 and used by practitioners worldwide.
Software Active
Digital Content Protector
Deployable verification of watermarks in AI-generated and AI-edited content, developed with CSIRO Data61 to help platforms and users tell authentic media from manipulated media.
Interested in collaborating on a demonstrator or funded project? See industry and government partnership.