Every AI model you use has real people staring at the darkest, most disturbing content — child sexual abuse material, suicide prompts, extreme violence — labeling it and teaching the model what it is and how to respond.
This job is called data labeling.
Among the companies that handle this work, Scale AI is one of the largest.
What Scale AI does
Scale AI was founded by Alexandr Wang when he was 19. It is now one of the most important companies in AI data infrastructure. Its clients include OpenAI, Google, Meta, and the U.S. Department of Defense.
In 2024, Meta invested $14.3 billion directly, becoming a major shareholder in Scale AI.
This means Scale AI is now partly owned by Meta while continuing to provide training data services to Meta — a peculiar loop.
Labor Department investigation
In March 2025, TechCrunch reported that the U.S. Department of Labor (DOL) was investigating Scale AI for potential violations of the Fair Labor Standards Act (FLSA).
The investigation was launched during the Biden administration and remained unknown to the public for nearly a year.
The focus is on Scale AI's workers, whom the company calls "Taskers." Scale AI classifies them as independent contractors rather than employees, a distinction with real legal consequences: contractors are not entitled to federal minimum wage protections, Social Security, or working hours protections.
The situation is more direct in the Philippines: local workers labeling data for Google's autonomous driving project have reported pay below the local statutory minimum wage, with no health insurance or paid leave.
Scale AI has denied the allegations of violations.
Schuster v. Scale AI: psychological harm case
Money is not the core issue.
A case is being heard in the U.S. District Court for the Northern District of California. The plaintiffs are former Scale AI "Taskers," and their central claim is that the company required workers to spend prolonged periods exposed to harmful content, causing severe psychological injury, and that the company failed to provide adequate protection.
These workers were required to engage in extended conversations with AI systems on topics including:
- Suicide prompts and methods
- Child sexual abuse material (CSAM) related content
- Extreme violence and predatory scenarios
According to the complaint, this was not quick, click-away work. The AI system would follow up on each worker response, demanding further detail and requiring workers to continuously engage with such content, "requiring complete mental immersion."
The complaint cites two concepts: moral injury — psychological damage from being forced to repeatedly act against one's own values — and institutional betrayal, referring to an employer knowing workers face risks yet failing to protect them.
Workers described specific symptoms including PTSD, depression, and anxiety.
The lawsuit alleges negligence and violations of California's Unfair Competition Law (UCL), and is seeking class-action status.
Not the first time
This pattern is not new.
In 2020, The Verge published an in-depth report exposing that Facebook (now Meta) employed content moderators who were heavily exposed to terrorism and child violence content, with workers commonly developing post-traumatic stress symptoms. Eventually, Facebook's contractor Cognizant settled related lawsuits for $52 million.
But that was content moderation — workers' task was to remove harmful content from the platform.
Scale AI's workers are doing AI training data labeling — they are teaching models "how to respond to this type of content." This work is more hidden, less publicly understood, but the degree of psychological harm is no less than content moderation.
Industry structural problem
The Scale AI case reveals a systemic arrangement in the AI industry: outsourcing the most harmful and difficult work to the lowest-paid workers, using "independent contractor" status to bypass labor protections, while talking loudly about "responsible AI" and "safety first" on product launch pages.
In 2024, Meta spent $14.3 billion to acquire a majority stake in Scale AI. During the same period, Meta published numerous statements and research papers on "AI safety" and "responsible AI." But the workers in the Philippines labeling harmful content every day, possibly earning below minimum wage, cannot find their names in any official statement.
The labor of these workers is the material foundation of AI capability, yet this cost never appears in the training cost comparison tables of large models.
Now, the courts are beginning to calculate that bill.
Sources: Scale AI is being investigated by the US Department of Labor (TechCrunch); Training Artificial Intelligence and Employer Liability: Lessons from Schuster v. Scale AI (Epstein Becker Green Workforce Bulletin); Scale AI, Data Labeling Company, CocoLoop, Under Investigation by the US Department of Labor (AIBase)