待翻译：AI fired an S.F. store employee. Will California crack down on 'robobosses'?

AI 服务暂时不可用，以下为来源正文，待恢复后补全翻译。

Last week, a San Francisco store announced a personnel change in a social media post. An employee at Andon Market was fired for habitual tardiness. It would be an otherwise unremarkable decision, if not for the person who had done the firing — since they don’t technically exist. Andon Labs, an artificial intelligence company that runs Andon Market in Cow Hollow, said it believes the decision marked the first time an AI boss fired a human employee. While Andon Market is an explicitly AI-run operation billed as an experiment, many worker advocates worry AI-led firings could become far more widespread. They’re backing a bill in the Legislature that would place limits on so-called AI robobosses. Senate Bill 947 from Sen. Jerry McNerney, D-Pleasanton, would place new restrictions on the use of automated decision making systems, requiring both the disclosure of their use and mandate human review when the decision to discipline or fire an employee is primarily based on automated decision-making systems, or ADS. The measure passed the Senate in June and cleared a key hurdle in the Assembly last week. “Nobody wants to be fired by a machine,” McNerney said. “Not without human beings at least looking at that record, at what grievances caused that separation.” Sometimes called algorithmic management, ADS are computer programs that analyze data to find patterns or correlations, according to a 2025 report from the UC Berkeley Labor Center. Researchers Annette Bernhardt and Lisa Kresge wrote in the report that the technology, which commonly employs AI, can be used in workplaces to give workers directions about their job tasks, predict workers’ future behavior and rank workers. The AI-run Andon Market in San Francisco attracted attention earlier this year when it opened, with the novelty of the AI boss, Luna, spurring both interest and apprehension. Andon Labs signed a three-year lease for the space on Union Street in the spring, with a specific goal: make an AI-run store profitable. Though human employees are at the gift shop, Luna handles all the top-level elements, such as hiring workers, setting schedules, managing merchandise selection and pricing and many other aspects one would expect from a retail boss. The store’s current collection is a melange of apparel, household goods and a few craft kits. T-shirts, baseball caps, coffee mugs and tote bags feature the Andon Market logo, sold alongside a watercolor kit, puzzles and postcards. The cofounders of Andon Labs, Lukas Petersson and Axel Backlund, have described the store as a type of experiment, pointing to it as a way to run AI systems against real-world problems and encourage conversations over the technology’s use in the workplace. “We are not taking a stance on whether or not we should have automated AI stores,” Backlund told the Chronicle. “We believe the models will get so good that there will be financial pressure on companies to adopt AI more and more, so we try to give AI responsibilities in a controlled setting as early as possible so we know what can go wrong and how AI behaves when it is a boss.” In Luna’s case, she started out, to Andon Labs’ own description, as a bit lackluster. While she had been running the day-to-day needs such as creating schedules and hiring new employees, her more granular management had some holes. In a report published on its website last week, Andon Labs said Luna had to be prompted to make basic employee rules. But even once a handbook was created, Luna forgot about it, and failed to take action after it logged the now-fired employee violating the lateness policy. Andon Labs wrote the policy had “simply vanished” from Luna’s memory. In the report, the company said Luna ended up suggesting that both Petersson and Backlund fire the employee in person, which the cofounders agreed on. But if McNerney’s bill passes in the coming weeks, that would no longer be a choice, and employers would also have to be transparent about the use of ADS to make critical personnel decisions. It would also ban employers from using ADS to predict behavior of their employees. If the technology leads directly to a decision to discipline or fire an employee, the bill would require employers to disclose its use to affected workers. The bill would also give employees in the state access to the data employers use to feed into the decision-making systems. If employers fail to abide by these requirements, employees would be able to contact the state labor commissioner to help enforce the law. Employers would be subject to state-led legal action and a $500 penalty per infraction. The biggest opponent of the bill, the California Chamber of Commerce, included it on its list of “cost-drivers” it is opposing in the Legislature this year, calling the requirements “impractical” and warning it will discourage the use of ADS tools. The business advocacy group said it could lead to costly penalties for businesses. The bill is sponsored by the California Federation of Labor Unions, AFL-CIO. In a news release upon the bill’s introduction earlier this year, the federation’s president, Lorena Gonzalez said there needs to be restrictions on how employers can use artificial intelligence to discipline and fire workers. “Employers are devastating workers’ livelihoods and taking no responsibility for the callous decisions of this unchecked technology,” Gonzalez said. “This is unacceptable. We need stronger guardrails to make sure there is human review and oversight of any decision made by a machine that impacts a worker’s job and paycheck.” But SB947 is not the first attempt to regulate these AI bosses. McNerney, who has authored a handful of AI bills over the past two years, first attempted to implement these regulations in Senate Bill 7 last year. Lawmakers approved it, but the bill was vetoed by Gov. Gavin Newsom. Newsom called the bill “overly broad” in his veto message and warned it duplicates some elements of existing law. This year’s bill, however, has made a few crucial changes addressing other misgivings from the governor, including leaving out a provision that would have regulated the use of customer rating data in ADS systems. “I’m confident that the bill will be passed out of the Assembly,” McNerney said. “I’m not quite sure where the governor is on this yet. But we’ll have to wait and see.” The bill will need to clear both houses of the Legislature by the end of the month. The governor will have roughly the month of September to make his final decision on the bill. McNerney stressed he’s not trying to ban the technology outright, but to ensure there is adequate oversight of AI as it continues to revolutionize many industries. “AI is going to be presenting harms, and we need to put standards in place to help protect people,” he said. Petersson and Backlund said they don’t have a position on the bill, and are in favor of what they call “democratic discussions” about the use of AI in the workplace. Yet they both questioned whether human review would be able to keep up with AI models as they rapidly evolve. “It might not be enough to just review the actions of the AI, because the AI will be smarter than the humans,” Petersson said. “There will be no point in putting in any effort reviewing it, so then the question is, how do you make a future-proof bill? We don’t have the answer to that.”