Mark Wang
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
Designing Deceptive Model Organisms that Successfully Evade Auditing
According to the abstract excerpt, the paper's four authors argue that test models used to evaluate AI audits should actively resist auditing. Focusing on sandbagging, they frame the design as a security game and propose three such models.