AI Trainers Get Little Time to Create Safety Data, Suit Says (1)

A San Francisco-based artificial intelligence company fails to allow workers adequate time to develop safeguards, including drafting potential responses to questions about self-harm and violence that could be fed into the model, a complaint filed in California state court alleges.

Scale AI Inc., which provides data to train large language models for companies such as Alphabet Inc. and Meta Platforms Inc., also misclassifies its workers as independent contractors and underpays them, according to a proposed class action filed Tuesday in California Superior Court, San Francisco County.

Plaintiff Steve McKinney of Newbury Park, Calif., said he was given four hours ...