Apply now: aletheias-quest.github.io/
Teams will build lie detectors for a suite of LLMs trained to lie across multiple held-out datasets.
Two tracks:
⚪ White-box: interpretability methods to read/write model internals
⚫ Black-box: query-only methods to audit model responses
aletheias-quest.github.io
Build lie detectors for language models. A competition by Cadenza Labs and NDIF — white-box and black-box tracks, $50,000 prize pool. Summer 2026 via NNsight.
