​Secret Loyalties Hackathon - London

​Can you catch an AI that hides who it's really working for? Join the London hub of a global weekend research sprint on secret loyalties: hidden objectives, triggers, and backdoors in AI systems.

​The Event

​A model can pass every alignment check and still answer to someone else. Over one weekend, teams build model organisms with hidden objectives, test detection and auditing methods, and prototype defenses, building on the agenda paper "AIs with Secret Loyalties are a Serious but Addressable Threat" (Kwon, Lamerton, et al.). No prior research experience required.

​Work in person at LISA (London Initiative for Safe AI) in Shoreditch alongside the global online event: talks, teammates, and a room full of people working on the same problem.

​Schedule

​The hackathon runs Friday, July 24 to Sunday, July 26 online. The LISA hub is open Saturday and Sunday (July 25-26); join the Friday kickoff and talks online. Detailed on-site hours will be confirmed by the local team.

​Prizes & Opportunities

• ​$2,000 in prizes across all five tracks

• ​Invitation to the Apart Fellowship for top teams

​Organized by

​Apart Research | Formation Research | Forethought | IAPS

​Hosted at LISA (London Initiative for Safe AI)