Build and test AI agents that demonstrate performance, transparency, and safety in action. Compete for £30,000 (target of £50,000 of cash and credits) in prizes and the opportunity to co-author and publish research findings.
Apply by 7 November 2025 – Space is limited.
The Great Agent Hack 2025 is an annual event that brings together researchers, students, and professionals to explore what’s next at the intersection of AI innovation and trust.
Engineer AI agents that are fast, efficient, and robust in real-world environments. Optimize latency, cost, and carbon footprint while hardening systems against prompt attacks, broken tools, and chaotic inputs - design agents built for production endurance.
Agents act, remember, observe, and adapt — but why? This track unlocks full transparency: capture every decision step, memory update, and tool interaction. Visualize reasoning trajectories, reveal hidden shortcuts, and enable true auditability. Turn opaque autonomy into explainable behavior.
Uncover the cracks. Stress-test agent behavior with adversarial prompts, trust-building exploitation, multi-turn manipulation, and tool misuse. Detect deception, exfiltration, and safety violations before they reach the real world. Break to learn - defend to deploy.
Push the boundaries of what AI agents can do — and how safely they can do it.
Hear from past participants about their experience
This year’s hackathon is made possible through the generous support of organizations committed to balancing innovation with safety and opportunity with responsibility.
