The paper "ContextBias" introduces a controlled evaluation framework to assess whether stereotypical associations learned by text-to-image models