Two bills on Newsom's desk would ban agents from firing you alone and give you a human in 15 minutes

California's legislature closed its session on August 31 by sending Governor Newsom two bills aimed squarely at the two places autonomous agents now touch ordinary people: the decision to fire them and the phone call when something goes wrong. Both were enrolled on September 4. Newsom has until September 30 to act. Neither has been reported outside employment-law alerts and bill trackers, which is odd, because between them they would set the first hard operating rules for agentic software in the largest state economy in the country.

SB 947: no firing by algorithm alone

Senate Bill 947, by Senator Jerry McNerney and carried in the Assembly by Ash Kalra and Chris Ward, is the bill its supporters call the No Robo Bosses Act. It adds a new part to the Labor Code governing what it calls automated decision systems: "any computational process derived from machine learning, statistical modeling, data analytics, or artificial intelligence that issues simplified output" and "is used to assist or replace human discretionary decisionmaking and materially impacts natural persons." The core rule is one sentence: "An employer shall not rely solely on an ADS when making a disciplinary or termination decision."

Where a system is the primary basis for such a decision, the bill requires the employer to "direct a human to corroborate the decision using data that was collected or used to produce the ADS output or other relevant corroborating or supporting information," which the text lists as supervisory evaluations, personnel files, work product, peer reviews or witness interviews. If the human cannot corroborate it, the employer cannot use the output. The worker must then receive "a written postuse notice at the time the employer informs the employee of the decision," saying that an automated system was primarily relied upon and that a human reviewed it. Employers are also barred from using such systems to infer a worker's protected characteristics or to penalize legally protected conduct.

Enforcement runs two ways. The Labor Commissioner can investigate and issue citations carrying a civil penalty of "$500 per violation." Workers also get a private right of action, with the bill authorizing injunctive relief, "punitive damages, and reasonable attorney's fees and costs." The Senate concurred in Assembly amendments 28 to 10. It takes effect July 1, 2027.

AB 1609: a human within 15 minutes

Assembly Bill 1609, by Assemblymember Rick Chavez Zbur, regulates the other end of the agent economy: the customer-service bot. It applies to any "large private business," defined as one "with more than five hundred million dollars ($500,000,000) in gross annual revenue nationally that provides goods and services to customers." Such a business must give "a clear and conspicuous disclosure that the customer service chatbot is artificially generated and not human" whenever "a reasonable person interacting with the customer service chatbot is likely to be misled to believe that the person is interacting with a human."

The sharper requirement is the service level. Every customer-service platform, including chat and telephone, must offer "a simple customer service feature" that lets a customer ask for a person, and the business must then make a good-faith effort to connect them within 15 minutes or "schedule an appointment time within one business day of the request at which a customer service agent will call or connect with the customer." Violations carry a civil penalty of up to $5,000 for a first offense and $10,000 for each one after, enforced by public prosecutors. Hospitals' health-care communications, consumer reporting agencies and regulated utilities are carved out. The Senate passed it 29 to 9 and the Assembly concurred 58 to 18.

What the vendors are about to inherit

Read together, the two bills draw a line that the agent industry has so far drawn only in marketing copy. An agent may recommend a firing but may not execute one alone, and the human who signs off has to have looked at something other than the agent's output. An agent may answer the phone but may not be the only thing that ever answers it. Both rules are cheap to state and expensive to build, because they require the human fallback to exist, be staffed and be reachable on a clock.

The customer-service rule lands hardest on the vendors who sell per-resolution pricing, where the economics depend on the bot closing the ticket without a handoff. A 15-minute human guarantee for any customer who asks is a cost line those models do not currently carry. The employment rule lands on the HR platforms that have spent the year putting agents into the talent function, and it arrives with a private right of action, which in California means it will be litigated. Neither bill is unusual for the state. California has repeatedly used its market size to set national defaults, and companies that redesign a product for Sacramento rarely maintain a separate version for everyone else. xAI's suit against Minnesota shows the industry is willing to fight state AI laws in court; these two, aimed at conduct rather than models, will be harder to challenge.

Newsom vetoed high-profile AI safety legislation in 2024 and signed a narrower version in 2025, so his signature is not automatic. But these bills are not about frontier models or catastrophic risk. They are about whether a person gets fired by software and whether a person can reach a person. Those are the questions agentic AI as a service was always going to have to answer. California has now written down two answers and put a price on getting them wrong.