Recommended Free Tools
AI systems are not autonomous in the way their interfaces suggest. People label training data, review disturbing material, test model outputs, handle logistics and keep outsourced operations running. Investigations and research describe low or disputed pay, unstable contracts, opaque management, weak worker voice and responsibility gaps between clients, vendors and subcontractors. Those findings concern particular projects and companies—not every AI workplace—but they show how an industry can hide human costs behind a polished product.
The human work hidden behind “artificial” intelligence
Fairwork’s AI research identifies annotation, content moderation, model evaluation and warehouse logistics as forms of labor that build and operate AI systems. The work is often distributed across countries and routed through labor platforms, staffing firms and subcontractors. That distance can make it difficult for a client, regulator or customer to see who performed the task, under what contract and with what protections.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Handbook of Labor Economics (Volume 6) | $149.99 | Buy on Amazon |
| 2 |
|
Labor Economics ISE | $55.50 | Buy on Amazon |
| 3 |
|
Modern Labor Economics: Theory and Public Policy (12th Edition) | $45.47 | Buy on Amazon |
| 4 |
|
Labor Economics | $90.85 | Buy on Amazon |
| 5 |
|
Black Labor, White Wealth : The Search for Power and Economic Justice | $23.93 | Buy on Amazon |
Fairwork summarizes the central contradiction this way: “There is nothing ‘artificial’ about the immense amount of human labour that builds, supports, and maintains AI systems.” Outsourcing is not automatically abusive, but fragmented chains can make accountability unclear when pay, workload, safety or dismissal becomes a problem.
What “treated badly” means in the evidence
The documented concerns are structural, not a single allegation about every worker. The same questions recur when comparing an AI company, a vendor or a platform:
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
| Issue | What to examine | Why it matters |
|---|---|---|
| Pay | Hourly or piece-rate pay, unpaid qualification and waiting time, deductions, targets and overtime | A client’s rate to a vendor is not the worker’s take-home pay. |
| Contract stability | Contract length, notice, removal from a project and what happens when a client leaves | A worker can lose income even when the end of a project is outside their control. |
| Exposure and support | Disturbing material, workload limits, counseling and time away from sensitive tasks | Safety measures must address psychological as well as physical risk. |
| Management and monitoring | Productivity metrics, cameras, software tracking, data collection and appeal processes | Opaque measurement can intensify pressure and reduce autonomy. |
| Voice and remedy | Grievances, representation, organizing rights and protection from retaliation | Without a credible route to challenge decisions, formal policies may have little force. |
| Responsibility | Which party controls pay, schedules, data, safety and subcontracting | Several companies can otherwise shift blame to one another. |
Fairwork’s framework is useful for asking these questions, but a score or certification cannot describe every worker’s experience or prove that conditions remain unchanged.
The Kenya content-moderation case involving OpenAI and Sama
TIME’s 18 January 2023 investigation examined 2021 contracts in which Kenyan workers at Sama performed harmful-content detection for an OpenAI project. The workers reviewed text that could include graphic violence, abuse and other disturbing material. TIME reported that OpenAI paid Sama a contract rate of $12.50 per hour; it also reported worker earnings estimates that varied by role, targets and the source of the account, with some workers describing effective pay below $2 per hour. Sama disputed parts of TIME’s reporting and supplied different expected task and earnings figures.
Workers told TIME that the material caused distress and raised concerns about the adequacy of counseling and other support. The report included responses from both Sama and OpenAI. These are allegations and accounts about a specific project and period, not evidence that every content moderator or current OpenAI supplier faces the same conditions.
Rank #2
What the Fairwork assessment found at Sama
In its 2023 assessment, the Oxford Internet Institute said Fairwork used desk research, management interviews and worker interviews to assess principles covering pay, conditions, contracts, management and representation. Sama received 5 out of 10 in that first assessment.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Scan for outdated or missing drivers - takes under a minuteDriver Scan →The report recorded concerns about precarious contracts, excessive overtime, job strain and discriminatory management practices. One interviewed worker described seven-day weeks from 7:40 a.m. to 6 p.m. and unpaid overtime over three months. Fairwork also described engagement by the company with the findings; engagement is not the same as proof that all problems were resolved.
“It’s not acceptable that workers in the AI industry are subject to working conditions that put their health, wellbeing and financial stability at risk,” said Dr Funda Ustek Spilda, Fairwork senior researcher and project manager.
When a client leaves, the job can vanish
In April 2026, the Associated Press reported that Meta ended a major Sama engagement in Nairobi. Sama said layoff notices would affect 1,108 staffers. AP also described former moderators’ allegations of poor conditions and inadequate support. Legal claims mentioned in that report were ongoing, so those allegations should not be treated as court findings.
The episode illustrates a practical risk of outsourced AI work: employment can be tied to one client relationship. A worker may have little notice or bargaining power when that relationship ends, even if the work itself remains valuable to the industry.
Workplace surveillance extends the warning beyond AI contractors
The same management logic is spreading into ordinary workplaces. The U.S. Government Accountability Office’s 2024 report reviewed 217 public comments from 211 stakeholders submitted in May and June 2023. Commenters described computer-monitoring software, cameras, microphones, geolocation, tracking applications and wearables. Their views on productivity and well-being varied.
Rank #4
Reported concerns included distrust, lower morale, stress, anxiety, privacy risks, bias and possible chilling effects on organizing. Because GAO summarized stakeholder comments rather than surveying a representative worker sample or estimating causation, the report cannot establish how common any one effect is.
The International Labour Organization’s working paper AI systems @ work: a changing psychosocial work environment, dated 30 April 2026, examines surveillance, data-driven management, autonomy and mental and social well-being. It also asks whether existing occupational-safety approaches are sufficient for risks associated with AI-enabled management. The question is broader than content moderation: an employer can use AI to allocate tasks, score performance or predict behavior without putting a worker anywhere near a training-data project.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why hidden labor matters to everyone else
Automation can shift risk instead of removing it
A system that appears automated may be moving difficult judgment, error correction and emotional strain to workers who are less visible and have less leverage. Customers then see speed and convenience while the supply chain absorbs the cost.
Best Value
- Used Book in Good Condition
Opaque metrics can become the default workplace
If constant tracking and target-based management are accepted in outsourced operations, similar practices become easier to justify in warehouses, call centers, offices and gig work. The issue is not whether measurement is ever useful; it is whether workers know what is collected, can challenge errors and retain meaningful control over their jobs.
Responsibility can disappear between organizations
AI developers, labor platforms, vendors and subcontractors may each control only part of a worker’s experience. Contracts that leave safety, pay or grievance handling to another party can produce a system in which everyone claims limited responsibility.
How to judge an AI job or supplier
Before accepting work or approving a vendor, ask for written answers to these questions:
- What is the worker’s actual rate, and are qualification, waiting and overtime hours paid?
- How long is the contract, what notice applies, and what happens when a project is canceled?
- Could workers encounter graphic or traumatic material, and who funds confidential, appropriate support?
- What monitoring tools and productivity targets are used, how long is data retained, and who can access it?
- Can a worker appeal a rejection, suspension or pay decision through a process independent of the immediate supervisor?
- Can workers organize or obtain representation without retaliation?
- Which company is legally and practically responsible when a subcontractor fails?
Ask for evidence rather than relying on ethical branding: sample contracts, pay records, escalation procedures, audit findings and named responsibility for subcontractors. Fairwork’s principles provide a starting framework, not a guarantee.
What credible accountability would require
- Traceable supply chains: AI clients should disclose the vendors and subcontractors performing core data, moderation and evaluation work.
- Worker-level pay transparency: Contracts should show the rate reaching workers and identify unpaid time, deductions and target penalties.
- Psychological safety: Sensitive-content work needs exposure limits, rotation, voluntary access to qualified support and a non-retaliatory way to stop unsafe assignments.
- Due process: Automated or manager-set productivity decisions should be explainable and appealable, with records workers can inspect.
- Shared liability: A client should not be able to outsource away responsibility for basic labor and safety standards.
- Independent worker voice: Representation and collective bargaining must remain possible across platform and subcontracting arrangements.
Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass by Mary L. Gray and Siddharth Suri offers broader context on invisible digital labor. It is background reading, not evidence about the specific cases above.
The bottom line
The strongest conclusion is not that every AI company treats every worker alike. It is that AI’s apparent automation depends on people whose labor can be outsourced, monitored and discarded with little public visibility. The Kenya investigation, Fairwork’s Sama assessment, the reported Nairobi layoffs and wider surveillance research show why fair pay, stable contracts, psychological protection, transparency and worker power must be designed into AI operations rather than added after harm appears.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
