Worker Discretion Advised: Co-designing Risk Disclosure in Crowdsourced Responsible AI (RAI) Content Work

Workers want meaningful choice, not just warnings: The study found that crowdworkers prefer disclosure mechanisms that give them genuine agency to make informed decisions about risky tasks, rather than simple warning labels that don't actually help them understand or avoid harmful content exposure.

Task designers face competing pressures: Those creating RAI content work tasks struggle to balance their need for worker participation with ethical obligations to protect workers from psychological harm—a tension that current platforms don't adequately support them in resolving.

One-size-fits-all approaches don't work: The research revealed significant differences in how individual workers want to receive risk information and what level of detail they need, suggesting that effective disclosure requires customization rather than standardized warnings across all tasks.

Platforms have largely abdicated responsibility: Platform representatives in the study acknowledged limited involvement in shaping risk disclosure practices, indicating a gap where workers and task designers are left to navigate worker protection issues without adequate institutional support or guidelines.

RAI work carries underrecognized harms: The study highlights that crowdworkers doing responsible AI tasks—like flagging harmful content or testing AI systems for bias—face similar psychological risks as content moderators, yet receive far less attention and protection in research and policy discussions.