California has enacted a package of measures designed to put limits around how employers use artificial intelligence in decisions that affect workers. The headline rule is refreshingly plain: an employer cannot rely only on AI when disciplining or terminating someone.

That does not amount to a ban on AI in the workplace, and it does not stop companies from using the technology to automate work or reshape their staffing. The measures instead focus on a narrower but highly consequential problem: what happens when software is involved in deciding who gets monitored, disciplined, relocated or laid off.

For workers, the difference is meaningful. An AI system may be used as part of an employment process, but it cannot be the sole decision-maker for a disciplinary action or termination. The new rules also require transparency where an AI system causes a mass layoff, relocation or termination. Separate protections address AI-powered surveillance in workplace bathrooms, while another measure prevents lawyers from fully outsourcing core legal work, including drafting briefs, to AI.

A guardrail against the automated firing screen

“AI” is an exceptionally broad label. In this context, it can cover software used to sort, rank, score or recommend actions based on employee information. An employer might use a system to identify workers for scrutiny, flag perceived performance concerns, or help choose candidates in a job cut. The concern is not merely that a machine has a role in the process; it is that its output can be treated as an unquestionable answer.

California’s new approach rejects that idea for discipline and termination. A person must remain involved rather than allowing an automated recommendation to be the sole basis for the outcome. That distinction matters because the practical effect of a tool is often defined by how much weight its recommendation carries. A manager who can genuinely review, challenge and override an AI-generated result occupies a very different role from one who simply clicks through a prebuilt list.

The legislation does not say AI cannot inform a workplace decision. It says an employer cannot make an adverse disciplinary or termination decision by relying only on it. That leaves room for AI-assisted processes while drawing a legal boundary around the most serious employment consequences.

What “human involvement” should not be confused with

The supplied details establish a prohibition on sole AI reliance, not a complete technical blueprint for every workplace review. Workers and employers should therefore avoid reading more into the rule than is currently clear. It does not, on the information available, establish that every AI-assisted employment decision is prohibited. Nor does it establish that the existence of a human name somewhere in a workflow automatically cures a flawed process.

Still, the policy direction is clear: a decision affecting someone’s job should not be reduced to an opaque automated score. That is particularly important when employees have limited visibility into the data fed into a system, the criteria it uses, or the assumptions built into its ranking.

Transparency becomes part of the layoff conversation

The measures also require companies to provide transparency if a mass layoff, relocation or termination is caused by an AI system. Transparency, in ordinary terms, means the affected process cannot remain entirely hidden behind a technical curtain. It puts AI’s role on the record when the technology is responsible for a significant employment action.

That requirement is distinct from the ban on sole AI reliance in individual discipline and termination decisions. One rule is about who or what may make a decision; the other is about disclosure when AI causes a broader employment event. Together, they acknowledge that AI can influence workforce changes at more than one level: it can shape decisions about an individual, or be implicated in a large-scale organizational action.

The available information does not specify the exact form, timing or audience for the required transparency. Those details matter in practice, especially for employees trying to understand why a decision occurred. But the underlying principle is already notable: when an AI system has caused a mass layoff, relocation or termination, its role is not something a company can simply keep offstage.

For businesses, this creates a practical governance issue. If an organization uses AI in workforce planning, it needs to know where the system’s recommendations end and where a human decision begins. If those lines are blurry internally, explaining the AI’s role externally will be harder. Clear records of how a system was used, what information it considered and who made the final call could become central to responsible compliance.

Why bias concerns are at the center of the issue

The policy arrives amid concerns that AI-based employment tools can produce unequal results. Complaints from Meta employees alleged that biased AI tools disproportionately selected people who had taken medical leave as layoff candidates. That allegation is a useful illustration of why employment software deserves scrutiny: even if a model is presented as efficient or neutral, it can produce results that appear to burden a particular group more heavily.

Bias in this setting does not require a system to be explicitly programmed with a discriminatory instruction. A system can draw conclusions from data patterns, proxies or historical decisions that embed disparities. If a process rewards signals that correlate with medical leave, for example, a supposedly neutral screening exercise may have a much more troubling real-world effect.

The new California measures do not, based on the supplied information, resolve every question surrounding biased algorithms. They do, however, intervene at two points where a bad automated output can do serious damage: the final decision to discipline or fire a person, and large-scale staffing actions caused by AI. Human review and required disclosure do not guarantee fairness on their own, but they create opportunities to identify errors, ask questions and assign responsibility.

Surveillance has a hard privacy boundary

The package reaches beyond layoffs. It also protects against the use of surveillance tools in workplace bathrooms. The point may sound obvious, but its inclusion is a useful reminder that workplace technology is not limited to spreadsheets and staffing dashboards. Monitoring tools can reach into physical spaces and daily routines, turning ordinary privacy expectations into a question of what an employer’s software can technically do.

The bathroom protection establishes a clear line: AI-powered surveillance does not belong there. It is a targeted rule, not a general description of every permitted or prohibited monitoring practice. Yet its significance is larger than the specific setting. It acknowledges that a capability is not automatically an acceptable workplace practice.

For employees, the key takeaway is that AI governance is also a privacy issue. For employers, it is a signal that deploying a monitoring system cannot be treated solely as an efficiency or security decision. Where, how and why a system observes workers matters.

One of the measures addresses the legal profession, prohibiting lawyers from fully handing core legal work over to AI, including the drafting of briefs. A legal brief is a written document submitted in support of a position in a legal matter. It is core legal work because it requires judgment about arguments, facts and the applicable law.

As with the worker-discipline rule, the language described here does not erase AI from the picture. The restriction is on fully handing over essential work. The practical principle is accountability: a lawyer cannot simply treat an AI system as the professional responsible for an argument that may affect a client or a case.

This parallel is notable. In both employment decisions and legal work, California’s measures focus on situations where automated output could be consequential but difficult to interrogate. The answer is not framed as “technology must never assist.” It is framed as “a responsible human cannot disappear from the chain.”

A state-level response as AI becomes routine management software

One in four managers reportedly uses AI to help decide which employees to cut often or all the time. Even without further detail about that figure, it points to the speed with which AI tools have entered managerial workflows. What might once have been a specialized analytics exercise can become a routine prompt, ranking or recommendation inside ordinary workplace software.

That is why rules around use matter before AI-assisted decision-making becomes invisible. A system does not need to issue a formal firing order to have power over someone’s employment. If its score determines which names receive closer review, or which employees are first placed on a reduction list, its influence can be substantial.

California is positioning itself as a leader in AI regulation. Governor Gavin Newsom said AI should expand opportunity rather than harm workers and families, and argued that public safeguards are necessary rather than relying on the industry to regulate itself. He has likened the technology’s potential danger to the public to that posed by the airline industry.

That comparison should not be read as a claim that AI and aviation function in the same way. It explains the regulatory posture: technology with broad public consequences may need rules, oversight and clear responsibility rather than a promise that voluntary restraint will be sufficient.

Newsom also signed an executive order directing state agencies to keep using the term artificial intelligence rather than super intelligence. It is a semantic decision, but language can shape policy debates. “Artificial intelligence” describes the technology category at issue in the laws; more expansive terminology can imply a different and more speculative frame.

What workers and companies can take from the measures now

For workers, the central practical point is that an AI system should not be the only basis for discipline or termination. If AI appears to have played a role in a mass layoff, relocation or termination, the legislation’s transparency requirement becomes relevant. The measures also place a specific privacy protection around workplace bathrooms.

For companies, the compliance challenge is less about having a single AI policy on a shelf and more about mapping real processes. Which systems rank or score employees? Are managers using those outputs when considering discipline or termination? Can the organization identify when a mass employment action was caused by AI? And are sensitive surveillance uses clearly blocked?

  • Do not let an AI output become the sole basis for discipline or termination.
  • Know when AI has caused a broad workforce action so required transparency is possible.
  • Maintain a meaningful human role where decisions carry professional or employment consequences.
  • Recognize privacy limits, including the protection against AI-powered bathroom surveillance.

The wider technology conversation continues to move quickly, from workplace software to increasingly capable models such as those discussed in Google’s Gemini 4 Argon claims. California’s new measures are a reminder that the important question is not only what an AI system can do. It is also who bears responsibility when its recommendations reshape someone’s livelihood, privacy or legal rights.