Brian Seibert's New Yorker piece on Buddy Bradley is ostensibly a book review, but it lands as a diagnosis. Bradley was a Black choreographer who taught white stars the latest moves in the 1920s and 1930s, received almost none of the credit, and has been largely erased from dance history. This week's arXiv paper on auditing African content moderators' working conditions using GDPR frameworks describes a structurally identical system: workers in Kenya and Nigeria performing invisible labor that makes AI systems functional, with minimal protections, minimal credit, and minimal recourse.

The Architecture of Erasure

Bradley's erasure was not accidental. It was produced by a set of conditions: racial hierarchy, contractual ambiguity, and an industry that preferred to credit the performer over the teacher. The content moderation industry replicates this with algorithmic precision. A 2023 paper in Big Data and Society by Miceli, Posada, and Yang found that data annotation and content moderation workers in the Global South are systematically excluded from the provenance documentation of AI systems they help build, creating a form of structural attribution erasure that is baked into the model training pipeline itself.

Where Credit Goes to Die

The GDPR audit paper is notable because it tries to apply a European legal framework to a labor condition that exists specifically because it was offshored to avoid European-style protections. Anthropic, which is simultaneously expanding to small business customers, will be selling tools whose outputs depend on this invisible labor chain. The Bradley case shows that erasure is durable: Maureen Footer's biography arrives a century after the fact. The content moderation audit is trying to intervene in real time, using the regulatory tools available, which is both more urgent and more likely to be ignored. , which makes the attribution gap a due diligence question as much as an ethical one.