AI Control Hackathon - Sydney

Three days building defenses for AI systems that might work against you, guided by Redwood Research and judged by safety orgs doing the real work.

Participants collaborate over three days to develop and test AI control protocols, evaluation tools, and safety measures designed to keep AI systems safe even when they may be working against human interests. Teams work with infrastructure like ControlArena and SHADE-Arena to stress-test existing safety measures and create novel control solutions. Top performing teams receive prizes, fellowship opportunities, and conference tickets with travel support.