Published: 26 February 2026
What happened: The Guardian collected first-hand accounts from workers across editing, translation, marketing, healthcare, and mathematics who have been paid to train AI systems — and are now watching those systems displace or undercut their roles.
Why it matters: The accounts reveal a consistent pattern: workers recruited to improve AI accuracy find their pay cut, their roles reduced, or their jobs eliminated shortly after completing the training work — raising sharp questions about transparency and consent in corporate AI deployment.
Wider context: IMF analysis cited in the piece estimates AI will affect around 40% of jobs globally. One contributor describes the experience as "digging your own digital grave" — a sentiment that cuts across industries from publishing to palliative care and pure mathematics.
Background: Experiences varied: a UK editor now earns less correcting AI-generated errors than she previously earned editing from scratch, and says it takes longer; a Milwaukee marketing writer was laid off two weeks after handing in AI workflow documentation his employer now uses to direct junior staff.
Keen bosses, strange mistakes and a looming threat: workers on training AI to do their jobs — The Guardian
Singularity Soup Take: There is something uniquely corrosive about a model that extracts workers’ expertise to automate their own roles — doing so without disclosure, as in Christie’s case, crosses an ethical line that no productivity argument can comfortably excuse.
Key Takeaways:
- Pattern of Deception: Several workers reported being asked to train AI systems without being told that was the purpose — including an editor who discovered months later that the “assistant editors” she was correcting were AI models, not people.
- Pay Cuts for More Work: At least one worker now earns less correcting AI-generated errors than she previously earned editing from scratch, describing the result as both slower and lower quality than doing the work herself without AI involvement.
- The Documentation Trap: A marketing writer was laid off just two weeks after completing six months of AI workflow documentation; his former workload is now carried out by junior employees following the AI prompting guidelines he wrote.
- Long-term Uncertainty: A French mathematics professor believes AI will fundamentally transform his field within a decade, but notes his concerns would be far sharper if he were starting his career today rather than at 44 with institutional job security.
- Healthcare Edge Cases: A palliative care chatbot project in Cardiff found AI struggling with patient pronunciation, misspellings, and dialects — and required careful handling of distressing queries, including questions from patients about ending their own lives.