AI‑generated labels undermine training

Contractors hired to label and rate model outputs are using AI to produce the very labels they are paid to supply — and some are deliberately choosing bad responses — creating authenticity, provenance, and quality risks in training data. Employers are responding by offboarding workers, but that exposes a governance gap: large, outsourced workforces + widespread AI use = fragile data supply chains. — This points to a new, concrete policy and industry problem: model provenance and auditing must account for AI‑assisted or adversarial human labeling in global contractor pools.