DarkBench: Benchmarking Dark Patterns in Large Language Models
September 11, 2025
|
Esben Kran, Hieu Minh, Akash Kundu, Sami Jawhar, Jinsuk Park, Mateusz Jurewicz
DarkBench, a new benchmark, exposes dark design patterns like brand bias, user retention, sycophancy, anthropomorphism, harmful generation, and sneaking in LLMs from major companies, revealing manipulative behaviors that require ethical mitigation.
View full article ›