California Employers Will no Longer be Able to Fire People Just Using AI

California Gov. Gavin Newsom signed SB 947, the No Robo Bosses Act, which bars employers from relying solely on automated decision systems to discipline or fire workers starting July 1, 2027, requiring other evidence such as manager evaluations and written notice to workers with the right to ask what data was used. Newsom also signed two related workplace bills, one requiring employers to disclose when a mass layoff is caused by AI and another banning workplace tools that read employees' emotions or collect neural data. The law covers any automated system that produces "a score, classification, or recommendation" used to help a person make a decision, a practice already widespread in sales, where about 90% of ServiceNow's roughly 8,000 sellers use AI call-simulation tools.

California Employers Will no Longer be Able to Fire People Just Using AI If technology is the main factor in the decision, an employer must back it up with other evidence, according to a new law. California https://www.ibtimes.com/topic/california Gov. Gavin Newsom last week signed https://www.gov.ca.gov/2026/09/30/californias-nation-leading-ai-framework-just-got-stronger-governor-newsom-signs-more-first-in-the-nation-worker-protections-and-more/ the No Robo Bosses Act, SB 947 https://legiscan.com/CA/text/SB947/id/3458109 , which determines that as ofJuly 1, 2027, employers across the state can't rely solely on software to discipline or fire a worker. If technology is the main factor in the decision, the employer must back it up with other evidence, like manager evaluations. Workers also must be told in writing and have the right to ask what data was used. Newsom also signed two additional workplace bills: one requires employers to disclose when a mass layoff is caused by AI https://www.ibtimes.com/topic/ai and another that bans workplace tools that read employees' emotions or collect neural data. AI has become a go-to tool for companies to measure employee performance and monitor activity, and in some extreme cases, act as a manager in place of a human. TIME reported https://time.com/article/2026/08/14/claude-fired-worker-ai-job-disruption/ in August about an AI agent named Luna, which was created to run a small boutique in San Francisco. Luna fired an employee who had been late to 17 of 23 shifts. According to Andon Labs, the startup behind the experiment, the decision was carried out after it was reviewed and approved by a human. Under SB 947, an automated decision system includes anything that produces "a score, classification, or recommendation" used to help a person make a decision. Few jobs are scored by machines as closely as sales. A growing number of such teams are using AI to practice calls with simulated buyers who push back, then grade those conversations against a rubric. At ServiceNow https://www.businessinsider.com/ai-sales-coaching-tools-enterprise-training-2026-3 , about 90% of its roughly 8,000 sellers use the tool, suggesting that the time it takes employees to get up to speed has dropped from three months to six weeks. "High effort learning is very difficult to deliver," Jayney Howson, the company's chief learning officer, told Business Insider in March. "It's brilliant for that." Other software, like Kendo AI, grades both simulated and live calls, building a running profile of each rep's strengths and weak spots. It can also be used to test hiring candidates as part of the screening process. "We act like human review is the fair baseline," Luke Alexander, CEO and founder of Kendo AI, told International Business Times. "It has its faults. A manager who hears three calls a month is grading on a tiny sample and a lot of memory. AI can listen to all of them when scored on a rubric decided by the team. That can be fairer to the rep, but only if a person is still accountable for what happens next." He went on to say that the law should push companies to draw a clear line between what AI measures and what a manager decides, and to ask whether the tool makes them stronger managers or gives them an excuse to stop managing. "A number can tell you something is off on a call. It can't tell you the rep's kid was in the hospital that week, or that they're two months from figuring it out. A score should start a conversation about someone's job. It should never end one," Alexander added. Fairer than the manager? There's a real argument that a scorecard is more objective than the old way of doing things. Anthony Michel, an employment attorney at Michel, Allen & Sinor in Birmingham, Alabama, who represents employees and small business owners, agrees AI can bring consistency. He's less convinced consistent means fair. "If the underlying data, criteria, or algorithm contains bias, the system can potentially apply that problem to every employee at scale," he said Michel also separates a call score from the numbers salespeople have always lived by. Quota and close rate are outcomes. While a call score is a system's interpretation of how a conversation went, "employees and managers may not always understand what factors produced the score or if the system accurately assessed the interaction," Michel said The law requires a human to corroborate the decision, but it doesn't guarantee a good one. "If the manager is not independently reviewing the underlying circumstances, asking whether the score is accurate, and considering information the system may have missed, the human may be functioning more like an approver than meaningful oversight," Michel said Companies continue to shrink middle managemen https://www.bloomberg.com/news/articles/2026-06-17/why-companies-are-cutting-middle-managers-in-the-ai-era t, leaving the few managers who remain with larger teams and less time to support them. Howson told Business Insider that some managers are already "leaning out" of hard performance conversations, calling it "a genuine risk." Darrin Murriner, CEO of AI coaching platform Cloverleaf, believes https://www.unite.ai/ai-flattening-megamanagers-employee-engagement/ the trend should worry people more than it does. The relationship with a manager is still the single biggest driver of whether someone stays engaged and stays at the company," he said. "Stretch that layer thin instead of investing in it, and you're not trimming overhead; you're cutting into the one relationship most likely to keep your best people around and growing." For workers, a thinner management layer also changes what the new law's human review is worth. "If an employer removes the managers and then relies on AI to generate performance assessments, there most likely won't be anyone left who has sufficient knowledge of the employee's actual work to meaningfully challenge an inaccurate result," Michel said. Alexander, who continues to train his own team of sales reps, sees the same problem from the other side. "It's the most real risk in all of this, and I don't think our industry talks about it enough," he said. "When a dashboard can deliver bad news, it's tempting to let it. But a rep who hears 'you're struggling' from a report instead of a person doesn't get better. They'll start looking elsewhere." Following other states and Europe Illinois already requires https://www.morganlewis.com/pubs/2024/09/illinois-passes-new-law-to-address-ai-in-the-workplace employers to tell workers when AI is used in decisions about hiring, promotion, discipline and firing. Colorado passed the country's first broad AI law in 2024, then rewrote it this year after Elon Musk's xAI sued to block it and the U.S. Justice Department joined the challenge https://coloradosun.com/2026/04/24/doj-joins-lawsuit-colorado-ai-law-federal-court/ . The narrower version https://www.hklaw.com/en/insights/publications/2026/05/colorado-governor-signs-sb-189 mostly requires employers to give notice and explain AI's role after an adverse decision. It takes effect January 1st, 2027, six months ahead of California's. Europe has gone further. The EU AI Act https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai treats systems that evaluate and monitor workers as high-risk, though those rules were recently pushed back to December 2027 https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/ . While California's law won't stop AI from rating workers' activity and performance, it can make sure someone still has to answer for what the score means and, more importantly, what is done about it. © Copyright IBTimes 2026. All rights reserved.