techpolicy.press web signal

Ahmad: AI's 'outsourcing maze' hides who really controls work

TL;DR

  • Meta's termination of its contract with Sama left more than 1,000 Kenyan data workers redundant, a case study Ahmad uses to open her argument.
  • Cognizant exited its Meta content moderation project after Arizona conditions were publicly revealed, but transferred a dominant share of its Hyderabad workforce to another large subcontractor.
  • Existing supply chain due diligence laws only partially apply to AI companies, leaving regulatory voids around who controls subcontracted labor.

A new essay in Tech Policy Press from Weizenbaum Institute researcher Sana Ahmad reframes the recent wave of AI labor scandals as something more structural than bad-apple contractors. Ahmad opens on a familiar headline, the sudden redundancy of more than 1,000 Kenyan data workers after Meta terminated its contract with the data labeling firm Sama, and argues that this kind of episode is not a governance failure to be patched but the design of the AI labor economy working as intended.

The framing device is a phrase Ahmad borrows from South African journalist Marché Arends, the 'outsourcing maze.' Global AI firms sit at the top, contract to large subcontractors like Sama or Cognizant, who in turn hire the actual workers. Ahmad's argument is that this layered structure is not accidental. It lets lead firms buy labor without being seen to employ it, and it lets subcontractors exercise what she calls 'extensive normative and bureaucratic control,' cultivating cultures of professionalism, ensuring productivity, and reducing labor resistance, while shielding the client from view. Workers, she writes, are kept in the dark about the identity of the global firms they work for, the scale of the supply chain they are embedded in, and the source and location of the control exerted over their labor.

The Cognizant case does a lot of the essay's work. When negative publicity around working conditions at Cognizant's moderation center in Arizona pushed the company to exit its content moderation project with Meta, a dominant share of its Hyderabad workforce was transferred to another large subcontracting firm. The lead client changed the vendor. The workers changed the badge. The system kept running.

The honest caveat is that this is a single-author think piece grounded in named case studies rather than fresh field data, and Ahmad herself frames the failure as 'not an aberration but reflects the uneven development inherent to capitalism,' which is a broader claim than any one Meta contract can settle. What the reporting doesn't give you is a look inside the successor firm now holding those Hyderabad workers, or specifics on which supply chain due diligence laws might close the gap Ahmad flags, only that existing ones apply to AI companies partially at best.

For anyone building AI procurement or trust-and-safety functions, the takeaway worth sitting with is that vendor churn is not the same as a fix. Changing which subcontractor holds the contract can leave the labor conditions, and the reputational exposure, largely intact.

Shared on Bluesky by 3 AI experts