LCB-Bench: Benchmarking Large Language Models for Cognitive Bias Detection

We introduce LCB-Bench (Large Language Model Cognitive Bias Benchmark), a systematic evaluation framework for assessing LLM ability to detect cognitive biases in text. Our benchmark tests 30 cognitive biases across 7 categories using 515 bias-injected scenarios with controlled difficulty levels. We evaluate five frontier LLMs and find significant variation in detection capability across bias categories, with all models showing particular weakness in detecting social and group dynamics biases.