ContextBias introduces ContextBench, a benchmark of 92 professional roles and 1,656 controlled prompts, to test whether stereotypical associations between professions and visual attributes persist when placed in different contextual settings. Evaluating four state-of-the-art text-to-image models across 66,240 generated images, the authors find that placing a role in a semantically unrelated context does not suppress role-linked attributes — demographic cues and role-specific visual elements remain robust across contexts.
