Human Contractors Are Seeing All Your Horny - and Creepy - AI Prompts

Key Points:

• Hundreds of human contractors hired to improve Microsoft's Copilot AI are reviewing users' image editing prompts and photos, many of which contain dark and sexually explicit content, including nonconsensual and disturbing requests involving women, children, and cartoon characters.
• Contractors are instructed to evaluate how well the AI fulfills prompts using their "intuition," without judging the ethics of the content, leading to exposure to traumatizing material such as upskirt photos, pro-anorexia content, and images depicting animal sacrifice.
• The hidden human element in AI development, where workers review sensitive user data, is not clearly communicated to the public, raising privacy concerns as users often assume their interactions are private unless flagged for moderation.
• Workers report emotional distress and question the nature of their work due to the prevalence of sexualized and inappropriate prompts, highlighting issues with the ethical oversight and support for these contractors.
• This situation mirrors similar practices at other AI companies like OpenAI, where hundreds of employees also review user prompts to improve chatbot responses, underscoring broader industry challenges around content moderation and worker welfare.