California's No Robo Bosses Act Puts Human Review to Test

California Gov. Gavin Newsom signed SB 947, the No Robo Bosses Act, on September 30, putting a human reviewer between an automated system and certain decisions to discipline or fire a worker. The law bars employers from relying solely on an automated decision system for those decisions. When an employer primarily relies on one, a person must corroborate its output with supporting information. The provisions are scheduled to become operative July 1, 2027.

The title suggests a machine being escorted out of the manager’s office. SB 947 allows automated systems to inform decisions. Its safeguard turns on what happens after a system produces an output: whether a reviewer examines evidence that can support it, and whether the employer can use an output the reviewer finds misleading. A manager’s presence in the workflow offers little protection if the manager sees only the machine’s verdict.

The employee also gets information after automation plays a primary role. Under the notice provisions, an employer must tell an affected employee in writing when an automated system played a primary role in a disciplinary or termination decision and that a human reviewed the result. The employee can request a description of their own data primarily used by the system. Notice may help someone ask how a decision was reached; it cannot, by itself, establish that the reviewer reached an independent judgment.

What the Reviewer Must Be Able to Do

SB 947 calls for corroboration with information beyond the automated output, including such records as manager evaluations or peer reviews. If the reviewer finds the output inaccurate, incomplete or misleading, the employer cannot use it to make the decision. That gives the human role a consequence: a reviewer’s adverse finding cannot simply be passed along with an approval attached.

Consider a hypothetical tool that flags an employee for repeated missed assignments. A reviewer who receives only a risk score has no basis for checking whether the assignments were late, reassigned or recorded incorrectly. Access to the underlying records and relevant personnel information would let that person test the flag against events the score purports to describe. This is an illustration of the information a reviewer would need, rather than a prediction about how employers will implement SB 947.

Frank Dickson of Dickson Research put his practical test this way: “The test is whether the human checked the evidence and had the power to say no.” His assessment of the corroboration requirement points to two implementation choices a legal command alone cannot settle for every workplace: what the reviewer can inspect and what authority that person has. An employer could name a human reviewer on an organizational chart tomorrow. Giving that person usable records and permission to reject a recommendation requires decisions about the system and the workplace around it.

Those decisions may reach whoever buys the software. If an employer cannot retrieve information needed to check a system’s output, training a manager to click more carefully will not fix the problem. Dickson advises employers to identify tools that score or flag employees, give reviewers authority to overrule them, and keep the information a reviewer examined. These are his proposed practices, rather than findings about what California employers already do. A notice telling an employee that a person reviewed the result supplies a starting question: what could that person inspect?

The Narrower Bill Newsom Signed

The route to SB 947 helps explain why its obligations concentrate on adverse decisions. State Sen. Jerry McNerney first introduced the proposal in 2025 amid a labor push for limits on automated management. Newsom objected to an earlier version’s broad advance-notification requirement, which would have alerted workers whenever a potentially relevant system was in use. In his explanation, he argued that it could reach even innocuous tools. After that resistance, McNerney removed the advance-notification provision and language that would have extended protections to gig workers; Uber and Lyft had criticized the gig-worker language, CNBC reported.

The resulting law retains notice tied to a decision in which a system played a primary role. For a worker, that is a different point of disclosure: finding out that automation contributed to a disciplinary decision offers a route to questions about that decision, while advance notice could have identified a tool before it affected work. For an employer, the revised approach avoids an advance alert about every potentially relevant tool. The removed gig-worker language also means this version does not extend its protections to those workers; it does not answer what other rules may apply to their work.

Newsom signed SB 947 alongside other worker measures. SB 951 adds a technology-displacement disclosure to California’s existing notice rules for qualifying mass layoffs, relocations or terminations when AI or other automated technology wholly or substantially replaces or automates jobs. The two bills act at different moments. SB 951 concerns disclosure about a covered workforce change; SB 947 concerns review of an automated output used in an individual disciplinary or firing decision. A layoff notice does not corroborate a particular employee’s performance flag, and a reviewed performance flag does not tell a workforce which jobs a company plans to automate.

An employer planning a covered technology-driven layoff must identify the job functions being automated under SB 951. An employer using a system as a primary basis for an individual disciplinary decision faces SB 947’s corroboration and notice requirements. Workers can use the first disclosure to understand a change in jobs; the second gives an affected person information about a decision concerning them. Neither mechanism substitutes for the other.

Robert Singleton of the Chamber of Progress urged Newsom to veto SB 947, arguing in a letter that “primarily relies” gives employers no objective line between a tool that informs a decision and one that supplies its main basis. He warned that uncertainty could discourage tools that improve consistency or identify safety risks. Employers deciding which systems require review have a reason to want a workable boundary. Singleton’s objection does not establish how that boundary will be applied.

For an employee whose manager receives an automated flag and then adds a short note, the boundary becomes concrete: how much of the decision rests on the flag? The corroboration requirement asks the reviewer to examine supporting information, while the phrase primarily relies determines when that requirement applies. Even a thorough review cannot protect a decision the rule does not reach. Where the rule does apply, the evidence available to the reviewer will determine whether the human can challenge the system’s conclusion or merely put a name beside it.