The responsible decision-maker
If a human name appears beneath a decision ending someone’s employment, that person must also be able to change it. California’s newly signed AI laws bring this question into the workplace: employers cannot base discipline or dismissal solely on AI. To me, the value lies in keeping responsibility with the employer, rather than adding another signature between a software recommendation and a lost job. A manager implementing an automated recommendation must remain responsible for the decision.[1]
Human oversight creates room to protect a worker, but reading a recommendation and questioning it are different tasks. My concern is a manager reduced to approving the output. If the recommendation is a starting point, review may catch an error. If it is treated as settled in advance, a person’s presence may change little. Requiring a human assessment can nevertheless create a basis for meaningful scrutiny. The worker’s interest lies where the decision can actually change, rather than in the review’s name alone.[1]
Automation’s benefit therefore has to be considered alongside authority over the worker. My standard is whether management can stop a wrong recommendation as well as obtain one quickly. California’s measure does not prohibit AI from assisting a disciplinary decision altogether; it bars sole reliance on it. That distinction separates using a tool from handing it decision authority. For the employee, the meaningful gain would be an employer unable to distance itself from responsibility simply because technology was involved.[1]
From notice to working conditions
Another provision requires disclosure when AI causes a mass layoff, relocation or termination. This puts the information burden back on the employer: workers need not be left to decipher the technology’s role entirely on their own. I see that disclosure as a possible starting point for discussing why work changed. The employer’s stated reason may still be only part of the explanation. Notice gives workers a rationale they can discuss; how that rationale relates to other factors deserves separate attention.[1]
The package also prohibits workplace surveillance tools in bathrooms. This concerns a different working condition from dismissal: it defines a place where an employer may not monitor staff. I do not think privacy can be judged solely by whether a measurement is accurate. Some measurements should not be made. That boundary matters because a worker is more than a continuous source of data. It is also a narrow boundary: a bathroom prohibition does not mean every form of monitoring elsewhere in the workplace has been removed.[1]
What would make these measures valuable to workers is how oversight, disclosure and privacy change ordinary decisions. My expectation is not that employers abandon automation, but that those choosing it retain decision authority and the burden of explanation. When a software recommendation is not treated as the end of the matter, a worker’s situation can be reconsidered. When the technological rationale is disclosed, changes to work become discussable. California’s signed package opens those spaces; workplace power depends on people being able to change decisions, rather than merely carry them out.[1]