Regulating Algorithmic Management

Employers now hire, schedule, surveil, rank, pay, discipline, and fire workers through systems the workers cannot see and cannot contest. The legal framework governing this conduct was built for human supervisors, not software. I will work to pass a federal Algorithmic Management Standard governing any automated system that materially informs a decision about a worker.

The European Union already treats employment and worker-management AI as high risk, subjecting it to mandatory risk assessment, data governance, logging, human oversight, and accuracy obligations, and it bans emotion-recognition systems in the workplace outright.¹ The United States must go further. Coverage cannot turn on a worker’s classification. The independent contractor label, used to escape minimum-wage, overtime, and bargaining obligations, would not exempt an employer from these duties, because the duties attach to the act of algorithmic control, not to a contested employment status.

Disclosure, Audits, and a Human in the Loop

Workers and applicants must be told, in plain terms, when an automated system is used in a decision affecting them, what data it relies on, and what it measures. Illinois now treats the undisclosed use of AI in employment decisions, and the use of proxies such as ZIP code, as a civil rights violation, with liability for discriminatory effect regardless of intent.² High-stakes systems must undergo independent, published bias audits before deployment and at regular intervals after it. No consequential decision, above all a termination, may be executed by algorithm alone.

Pay and Surveillance

Algorithmic wage-setting must be prohibited. Gig platforms increasingly use granular surveillance data, including acceptance rates, location history, and how little a given worker has been willing to accept, to calculate the lowest pay each individual will tolerate, so that two people doing identical work earn different, unpredictable, personalized rates. Human Rights Watch documented this across seven major US platforms: six of the seven set pay by opaque algorithm, workers did not learn their pay until after finishing a job, and surveyed workers in Texas earned roughly thirty percent below the federal minimum wage and about seventy percent below a living wage.³ Pay must be transparent, predictable, and tied to work performed rather than to a model of a worker’s individual desperation. Workers must also own, and be able to export, the performance and behavioral data these systems collect about them.

Workplace surveillance must be constrained at its source. Continuous tracking of location, communications, and biometric and behavioral signals chills the protected activity of organizing and turns the workplace into an instrument of control. Monitoring must be limited to defined, legitimate purposes, barred where it captures protected concerted activity, and prohibited from reaching off-duty conduct. Federal labor law already supplies a foundation: pervasive monitoring that interferes with the right to organize is an unfair labor practice, a position the National Labor Relations Board has begun to develop and that statute should make explicit.⁴

These federal rules would set a minimum standard, which states would remain free to build on, and a private right of action would run alongside public enforcement so that harmed workers can pursue their own losses directly. Worker-management systems belong within the high-risk class of a Federal AI Authority, passing through pre-deployment approval with the Authority setting the standards and existing agencies enforcing them under non-displacing interagency agreements. Unregulated, these systems do not manage labor. They violate the rights of the people who perform it.

---

References