alizk2389/secret-loyalties-hackathon

Detecting Principal-Conditional Behaviour Across Conversational Sessions Apart Research Secret Loyalties Hackathon · 24–26 July 2026 · Solo submission · Tracks 3 & 4Author: Alishba Zainab Khan · Independent AI Safety Researcher · BS Data Science & AI, BITS Pilani If surfacing a secret loyalty requires eliciting the harmful behaviour it is trained to produce, then safe black-box auditing may be structurally impossible for that loyalty class. The activation condition is the… See the full description on the dataset page: https://huggingface.co/datasets/alizk2389/secret-loyalties-hackathon.