PublicAI runs a decentralised marketplace for AI data work: collection, annotation and model evaluation across text, audio, video and mapping. It claims 3.5 million workers, 300,000 validators and 200,000 builders, with quality enforced by a validator network that stakes on the blockchain rather than by an internal QA team.
Staking is the mechanism worth examining. Crowd annotation has one structural failure – a worker paid per task earns more by answering fast than by answering well – and the usual fix is a sampled review by someone paid properly, which does not scale with the crowd. Putting a validator’s own stake at risk on the labels they approve attacks the incentive directly instead of inspecting the output. Whether it works at scale is the open question.
PublicAI Foundation, registered in the Cayman Islands, has raised $10 million. There is a free tier with a complimentary annotation allowance and pay-as-you-go overage, then PublicAI Pro for enterprise, with unlimited team members and no published figures – you pay for completed work only. A crypto-native quality layer adds a token economy to a data-labelling decision, distributed annotation across millions of workers makes consistent labelling guidelines the hard problem, and sourcing training data from social media content raises provenance questions the buyer inherits.
