Apple AI trainers reveal strict content rules and high-pressure working conditions

Finnish workers hired to train Apple’s artificial intelligence tools in Greece have described strict content restrictions and grueling working conditions, according to an investigation by Yle’s MOT program.

Trainers reported that Apple’s guidelines prohibited the AI from discussing controversial topics such as war, immigration, and abortion, with some calling the restrictions a form of censorship. One trainer stated bluntly: “I wouldn’t sell this crap to a customer.”

The trainers, who spoke anonymously due to strict confidentiality agreements with Apple, were employed through the outsourcing firm Teleperformance. Their role involved fine-tuning Apple’s AI responses to ensure accuracy and prevent harmful output. However, they described the work as repetitive and highly monitored, with software tracking their activity and productivity demands reaching up to 70 tasks per shift.

Some trainers admitted to bypassing surveillance by jamming keyboard inputs with objects like paper clips or toothpicks. Others resorted to using competing AI models—such as those from Google or Meta—to verify or generate responses, a practice that raises questions about data integrity.

Pay for the role was modest, around €1,300 per month, slightly below Greece’s average wage. The trainers also expressed ethical concerns, arguing that human oversight should be mandatory for AI responses. “From an ethical and moral standpoint, it’s essential that a human checks every answer,” one trainer said.

Apple has not publicly commented on the allegations. The company’s AI, still under development, is expected to launch in 2025. The trainers’ accounts shed light on the often opaque process of AI fine-tuning, where human labor in low-wage countries plays a critical but undervalued role in shaping high-profile tech products.