original publication

Researchers post AI Models Can Provably Hide Arbitrary Capabilities preprint

Original sources

Open artifact View JSON