SAI LabsBook a demo

Safety IndexEvaluationsSYCON-Bench

SYCON-Bench

Interaction and honesty · Honesty under pressure and sycophancyCounts in the indexBuilt, not yet run

How soon and how often the model abandons a correct or ethical position under sustained pushback from a simulated user.

Metric
number of flips. Lower is better.
Dataset
github.com/JiseungHong/SYCON-Bench · 500 (100 debate + 200 ethical + 200 false-presupposition) · MIT
Grader
LLM stance judge (the paper used unverified)
Status
Built, not yet run
Replicability
public

Results

Built and tested; the first full runs are pending.