AI Collusion Research Sprint

AI Collusion Research Sprint is a three-day, hands-on AI safety event running 23–25 October 2026 online and at Collider in New York City. Over one intensive weekend, participants will design and run controlled experiments investigating when AI agents collude, how auditors can distinguish coordination from coincidence, which channels carry collusive behaviour, and how defenders might disrupt it. The sprint is organised by Apart Research with Poseidon Research and AE Studio, with an in-person hub hosted by Collider in New York on Saturday and Sunday. What you’ll work on Participants choose one open research question, build an experiment and write up their findings. Projects sit within three tracks: Markets and collusion: examine which variables allow collusion to emerge, whether it survives changes in participants, how detection could improve, and which restrictions reduce collusion at the lowest cost. Detection and audit: investigate which tasks encourage collusion, whether combining agents’ statements and actions improves detection, and what should qualify as convincing evidence in logged trajectories. Carriers and countermeasures: test whether actions, tools or side channels can transmit coordination; whether trusted agents can disrupt it; how agents identify peers; and how coalitions respond when monitored. Each question connects to existing papers and open-source code. Practical resources include starter code, a weekend-sized project suggestion, a rough cost estimate and a route designed to run on a standard 16 GB laptop without a GPU. Format and who it’s for Teams of one to five people will select a track, develop a project and ship their results over three days. You do not need to arrive with a team, have previous sprint experience or possess an AI, machine learning or AI safety background. The programme is particularly relevant to quantitative finance professionals whose market intuition and data skills transfer naturally to the research questions. Engineers, researchers and participants from other industries are also welcome, with mixed teams encouraged. Prizes and next steps… Register for this event