Research preview. Independent expert review is pending. The worked analyses are hypothetical.
Method and corrections Interactive field model
Inspect the system around the AI system. The model is organized by the clauses of the definition: societies, the four objects, the benefits and harmful patterns governance aims at, and the conditions it races against. Apply a cross-cutting test, narrow by lifecycle stage or level, and inspect how definitions, claims, relationships and sources travel together.
Apply a cross-cutting test Whole model Power Legitimacy Distribution Adaptability
Narrow by lifecycle stage or level Lifecycle stage Any stage Research Data collection Model development Evaluation Procurement Deployment Monitoring Incident response Retirement Level Any level Community and organisational use Corporate Sectoral National Regional International Loading interactive controls. Concept links remain available. Field model · whole model
Groups follow the clauses of the definition; they are not a sequence or causal chain. Actual relationships appear in the inspector.
Jump to selected entity Societies …how societies should shape…
actor AI developer or provider actor Deploying organization actor Affected people actor Civil society and independent researchers Incentives …the incentives…
incentive Deployment pressure incentive Competitive race dynamics incentive Liability exposure incentive Procurement leverage
Institutions …institutions…
institution Regulator or public authority institution Internal safety and governance team institution Standards body institution International institution
Infrastructures …infrastructures…
infrastructure Evaluation capacity infrastructure Compute infrastructure Data, language resources, and data labor infrastructure Energy, water, and minerals infrastructure Data work and content moderation Safeguards …and safeguards around AI…
safeguard Risk management process safeguard Impact assessment safeguard Traceability safeguard Audit record safeguard Meaningful human oversight safeguard Supervision and regulatory review safeguard Pre-deployment evaluation and red teaming safeguard Incident reporting and post-deployment monitoring safeguard Complaint and appeal routes safeguard Review and sunset provisions safeguard Compute controls Benefits and harmful patterns …so that its benefits are broadly shared and harmful patterns are addressed…
principle Accountability principle Safety principle Broadly shared benefit pattern Contestable decision pattern Concentration of power pattern Systematically unreliable provenance pattern Over-reliance and deskilling pattern Sycophancy and engagement-driven agreement pattern Homogenisation and loss of authorship Choices and conditions …while they are still choices rather than conditions.
condition Lock-in condition Governance lock-in condition Reversibility window
Infrastructure · Infrastructures contested
Data work and content moderation The human labour that labels data, rates model outputs and moderates content so that AI systems work, often outsourced, low-paid and out of sight.
Who does the work that makes AI systems work, on what terms, and who can see it?
Tests Distribution Power Legitimacy
Lifecycle Data collection Model development Monitoring
Levels Corporate National International Recorded relationships Read each statement from its first entity to its second. Status describes our mapping, not certainty of cause.
No relationship has been recorded yet. This does not mean no relationship exists.
Claims and evidence · 4 source-assertion Confidence: medium Contestability: medium
An investigation found that Kenyan workers labelling harmful content for a chatbot’s safety filter took home roughly $1.32–$2 an hour while the contract paid the outsourcing firm $12.50 an hour; the firm disputed the take-home figures.
Our reading: One contract in one country, reported by journalists; the firm’s own figure was $1.46–$3.74 after tax. It shows how value is split along one chain, not the pay of all data workers.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic (opens in a new tab) Pay and contract paragraphs
Kenyan workers labelling harmful text for a chatbot safety filter through an outsourcing firm took home roughly $1.32–$2 an hour, while the contract rate paid to the firm was $12.50 an hour; the firm said workers could earn $1.46–$3.74 after tax. Workers described the work as psychologically taxing.
Scope note: Investigative journalism. The outsourcing firm disputed the pay figures; both ranges are recorded.
Passage checked 2026-10-18 · time-kenya-pay
source-assertion Confidence: high Contestability: medium
Independent ratings of 16 online labour platforms in 2024–25 found evidence that workers earn at least the local minimum wage on only four, and concluded that the platforms remain far from safeguarding basic fair-work standards; the best scored 8 out of 10.
Our reading: Scores reflect evidence the researchers could verify; platforms can dispute them.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Fairwork Cloudwork Ratings 2025 (opens in a new tab) pp. 3–4 and conclusion (p. 34)
Of 16 online labour platforms rated in 2024–25, evidence of policies ensuring the local minimum wage was found for only four; the authors conclude the platforms remain far from safeguarding basic fair-work standards and that unfair and insecure work is the norm on most.
Scope note: Research ratings of 16 online labour platforms against five fair-work principles; fieldwork February 2024 to January 2025. The final part of the report was not read.
Passage checked 2026-10-18 · fairwork-cloudwork-findings
source-assertion Confidence: medium Contestability: medium
Research with data workers in Latin America found that task instructions impose clients’ views on how data is labelled, and that precarious conditions make workers follow them rather than question them.
Our reading: Qualitative study of specific platforms and one outsourcing firm; it explains a mechanism rather than measuring its prevalence.
Inspect evidence · 1 passage Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
The Data-Production Dispositif (opens in a new tab) Abstract
Analysing 210 instruction documents and 55 interviews, the study finds that task instructions impose clients’ worldviews on data, and that precarious work and economic dependency make workers obedient to those instructions.
Scope note: Qualitative study of data workers in Venezuela and Argentina.
Passage checked 2026-10-18 · miceli-posada-abstract
interpretation Confidence: medium Contestability: medium
Data work is part of AI infrastructure: the pay, protection and voice of the people who label and moderate shape both who bears AI’s costs and the quality of the systems built on their work.
Our reading: AIGovSpec’s interpretation, placing labour inside the infrastructure object and the distribution test.
Inspect evidence · 3 passages Passage summaries are paraphrases, not quotations. Open the source to read the full passage.
Exclusive: OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic (opens in a new tab) Pay and contract paragraphs
Kenyan workers labelling harmful text for a chatbot safety filter through an outsourcing firm took home roughly $1.32–$2 an hour, while the contract rate paid to the firm was $12.50 an hour; the firm said workers could earn $1.46–$3.74 after tax. Workers described the work as psychologically taxing.
Scope note: Investigative journalism. The outsourcing firm disputed the pay figures; both ranges are recorded.
Passage checked 2026-10-18 · time-kenya-pay
Fairwork Cloudwork Ratings 2025 (opens in a new tab) pp. 3–4 and conclusion (p. 34)
Of 16 online labour platforms rated in 2024–25, evidence of policies ensuring the local minimum wage was found for only four; the authors conclude the platforms remain far from safeguarding basic fair-work standards and that unfair and insecure work is the norm on most.
Scope note: Research ratings of 16 online labour platforms against five fair-work principles; fieldwork February 2024 to January 2025. The final part of the report was not read.
Passage checked 2026-10-18 · fairwork-cloudwork-findings
The Data-Production Dispositif (opens in a new tab) Abstract
Analysing 210 instruction documents and 55 interviews, the study finds that task instructions impose clients’ worldviews on data, and that precarious work and economic dependency make workers obedient to those instructions.
Scope note: Qualitative study of data workers in Venezuela and Argentina.
Passage checked 2026-10-18 · miceli-posada-abstract