New Requirements for California Employers Who Use AI in Employment Decisions

TEXT START: The two new laws build on other recent AI-in-employment developments including updated regulations addressing the use of AI, machine learning, algorithms, statistics and other automated-decision systems used to make employment-based decisions.

The Dissection

This is not fundamentally an article about preventing AI displacement. It is a compliance memo that converts structural rupture into workflows: inventory the tools, document the inputs, insert a human reviewer, issue notices, preserve evidence, and amend vendor contracts.

Its accidental admission is more important than its legal checklist: AI is becoming part of the machinery that disciplines, terminates and replaces workers. The state is not restoring the employment–wage–consumption circuit. It is making the displacement legible, contestable and administratively orderly. The human reviewer is retained primarily as a legal control point, not because human labor remains economically necessary.

The Core Fallacy

The text confuses procedural control with substantive control.

Requiring human corroboration does not prevent the model from filtering evidence, ranking workers or defining the available choices. A human can become a compliant signature attached to an automated conclusion. A right to receive a description of personal data is disclosure, not ownership of the model, its objectives, its training data or the capital producing the decision.

Likewise, technology-displacement notices do not preserve jobs. They document the removal of jobs after the economic decision has already been made. Under P1–P3, legal friction may delay, tax or expose automation, but it cannot reverse it when automated systems remain cheaper, faster or more scalable and institutions cannot preserve human-only economic domains at scale.

The article mistakes an audit trail for a veto.

Hidden Assumptions

• Employers will accurately identify when AI substantially caused a layoff, despite incentives to describe decisions as ordinary restructuring or mixed human judgment.
• Human corroborators will exercise independent judgment rather than rubber-stamp an output already treated as authoritative.
• A meaningful data description will let employees challenge a decision, even when the causal chain is opaque, probabilistic or controlled by vendors.
• Penalties, litigation and attorney’s fees will outweigh the productivity gains from automation.
• State reporting can keep pace with rapidly changing tools, outsourced systems and hybrid human-machine workflows.
• Notice gives displaced workers meaningful leverage, retraining capacity or access to replacement demand.
• Individual claims and procedural rights can address an aggregate collapse in economically necessary work.
• Employers will absorb compliance costs rather than redesign workflows, outsource decisions, classify workers differently or move activity outside the most exposed legal perimeter.
• Preserving a human step preserves meaningful human agency.

These assumptions are not demonstrated. They are the load-bearing fiction that allows a transition-management regime to present itself as worker protection.

Social Function

Classification: transition management, ideological anesthetic and partial truth.

The partial truth is real: automated systems can produce discrimination, arbitrary discipline and concealed workforce displacement, and procedural rights can generate evidence useful in individual disputes.

But the dominant function is to domesticate the shock. The article replaces the question of who controls productive AI capital with narrower questions: Was a notice issued? Was a human involved? Was the data described? Was the vendor contract amended? That reframes dispossession as paperwork failure.

It also functions as a law-firm demand-generation memo. Systemic instability becomes an expanding compliance market involving audits, policies, investigations, vendor questionnaires, litigation and recordkeeping. Employers can then claim responsibility because they managed the paperwork surrounding the destruction.

The Verdict

This text confirms the Discontinuity Thesis more than it challenges it. California is not stopping the machine; it is requiring the machine’s owners to label the displacement, preserve a human signature and notify the people being removed.

The laws may create temporary lag, litigation niches and better evidence of the transition. They do not restore productive participation, mass wage income or worker control over AI capital. This is carcass management dressed as governance: an attempt to regulate the burial, not prevent the death.