After three years of trying to give Californians the right to know when AI is making a consequential decision about their lives and to appeal when things go wrong, Assemblymember Rebecca Bauer-Kahan said she and her supporters will have to wait again, until next year.
The San Ramon Democrat announced Friday that Assembly Bill 1018, which cleared the Assembly and two Senate committees, has been designated a two-year bill, meaning it can return as part of the legislative session next year. That move will allow more time for conversations with Gov. Gavin Newsom and more than 70 opponents. The decision came in the final hours of the California Legislative session.
Her bill would require businesses and government agencies to alert individuals when automated systems are used to make important decisions about them, including for apartment leases, school admissions, and, in the workplace, hiring, firing, promotions, and disciplinary actions. The bill also covers decisions made in education, health care, criminal justice, government benefits, financial services, and insurance.
Automated systems that assign people scores or make recommendations can stop Californians from receiving unemployment benefits they’re entitled to, declare job applicants less qualified for arbitrary reasons that have nothing to do with job performance, or deny people health care or a mortgage because of their race.
“This pause reflects our commitment to getting this critical legislation right, not a retreat from our responsibility to protect Californians," Bauer-Kahan said in a statement shared with CalMatters.
Bauer-Kahan adopted the principles enshrined in the legislation from the Biden administration’s AI Bill of Rights. California has passed more AI regulation than any other state, but has yet to adopt a law like Bauer-Kahan’s or like other laws requiring disclosure of consequential AI decisions like the Colorado AI Act or European Union’s AI Act.
The pause comes at a time when politicians in Washington D.C. continue to oppose AI regulation that they say could stand in the way of progress. President Donald Trump in July released an "AI Action Plan" calling for deregulation of the technology at the federal and state level. Earlier this year, Congress tried and failed to pass a moratorium on AI regulation by state governments.
When an automated system makes an error, AB 1018 gives people the right to have that mistake rectified within 30 days. It also reiterates that algorithms must give “full and equal” accommodations to everyone, and cannot discriminate against people based on characteristics like age, race, gender, disability, or immigration status. Developers must carry out impact assessments to, among other things, test for bias embedded in their systems. If an impact assessment is not conducted on an AI system, and that system is used to make consequential decisions about people’s lives, the developer faces fines of up to $25,000 per violation, or legal action by the attorney general, public prosecutors, or the Civil Rights Department.
Amendments made to the bill in recent weeks exempted generative AI models from coverage under the bill, which could prevent it from impacting major AI companies or ongoing generative AI pilot projects carried out by state agencies. The bill was also amended to delay a developer auditing requirement to 2030, and to clarify that the bill intends to address evaluating a person and making predictions or recommendations about them.
An intense legislative fight
Samantha Gordon, a chief program officer at TechEquity, a sponsor of the bill, said she’s seen more lobbyists attempt to kill AB 1018 this week in the California Senate than for any other AI bill ever. She said she thinks AB 1018 had a pathway to passage but the decision was made to pause in order to work with the governor, who ends his second and final term next year.
“There's a fundamental disagreement about whether or not these tools should face basic scrutiny of testing and informing the public that they're being used,” Gordon said.
Gordon thinks it’s possible tech companies will use their “unlimited amount of money” to fight the bill next year.
“But it’s clear,” she added, “that Americans want these protections — poll after poll shows Americans want strong laws on AI and that voluntary protections are insufficient.”
AB 1018 faced opposition from industry groups, big tech companies, the state’s largest health care provider, venture capital firms, and the Judicial Council of California, a policymaking body for state courts.
A coalition of hospitals, Kaiser Permanente, and health care software and AI company Epic Systems urged lawmakers to vote no on 1018 because they argued the bill would negatively influence patient care, increase costs, and require developers to contract with third-party auditors to assess compliance by 2030.
A coalition of business groups opposed the bill because of generalizing language and concern that compliance could be expensive for businesses and taxpayers. The group Technet, which seeks to shape policy nationwide and whose members include companies like Apple, Google, Nvidia, and OpenIAI, argued that AB 1018 would stifle job growth, raise costs, and punish the fastest growing industries in the state in a video ad campaign.
Venture capital firm Andreessen Horowitz, whose founder Marc Andreessen supported the re-election of President Trump, opposes the bill in part because it imposes costs "well beyond [California's] borders."
A policy leader in the state judiciary said in an alert sent to lawmakers urging a no vote this week that the burden of compliance with the bill is so great that the judicial branch is at risk of losing the ability to use pretrial risk assessment tools like the kind that assign recidivism scores to sex offenders and violent felons. The Judicial Council estimates that AB 1018 passage will cost the state up to $300 million a year. Similar points were made in a letter to lawmakers last month.
Why backers keep fighting
Exactly how much AB 1018 could cost taxpayers is still a big unknown, due to contradictory information from state government agencies. An analysis by California legislative staff found that if the bill passes it could cost local agencies, state agencies, and the state judicial branch hundreds of millions of dollars. But a California Department of Technology report covered exclusively by CalMatters concluded in May that no state agencies use high risk automated systems, despite historical evidence to the contrary. Bauer-Kahan said last month that she was surprised by the financial impact estimates because CalMatters reporting found that automated decisionmaking system use was not widespread at the state level.
Support for the bill has come from unions who pledged to discuss AI in bargaining agreements, including the California Nurses Association and the Service Employees International Union, and from groups like the Citizen’s Privacy Coalition, Consumer Reports, and the Consumer Federation of California.
Coauthors of AB 1018 include major Democratic proponents of AI regulation in the California Legislature, including Assembly majority leader Cecilia Aguilar-Curry of Davis, author of a bill passed and on the governor’s desk that seeks to stop algorithms from raising prices on consumer goods; Chula Vista Senator Steve Padilla, whose bill to protect kids from companion chatbots awaits the governor’s decision; and San Diego Assemblymember Chris Ward, who previously helped pass a law requiring state agencies to disclose use of high-risk automated systems and this year sought to pass a bill to prevent pricing based on your personal information.
The anti-discrimination language in AB 1018 is important because tech companies and their customers often see themselves as exempt from discrimination law if the discrimination is done by automated systems, said Inioluwa Deborah Raji, an AI researcher at UC Berkeley who has audited algorithms for discrimination and advised government officials in Sacramento and Washington D.C. about how AI can harm people. She questions whether state agencies have the resources to enforce AB 1018, but also likes the disclosure requirement in the bill because “I think people deserve to know, and there's no way that they can appeal or contest without it.”
“I need to know that an AI system was the reason I wasn't able to rent this house. Then I can at an individual level appeal and contest. There's something very valuable about that.”
Raji said she witnessed corporate influence and pushback when she helped shape a report about how California can balance guardrails and innovation for generative AI development, and she sees similar forces at play in the delay of AB 1018.
“It’s disappointing this [AB 1018] isn’t the priority for AI policy folks at this time,” she told CalMatters. “I truly hope the fourth time is the charm.”
A number of other bills with union backing were also considered by lawmakers this session that sought to protect workers from artificial intelligence. For the third year in a row, a bill to require a human driver in commercial delivery trucks in autonomous vehicles failed to become law. Assembly Bill 1331, which sought to prevent surveillance of workers with AI-powered tools in private spaces like locker or lactation rooms and placed limitations on surveillance in breakrooms, also failed to pass.
But another measure, Senate Bill 7 passed the legislature and is headed to the governor. It requires employers to disclose plans to use an automated system 30 days prior to doing so and make annual requests data used by an employer for discipline or firing. In recent days, author Senator Jerry McNerney amended the law to remove the right to appeal decisions made by AI and eliminate a prohibition against employers making predictions about a worker's political beliefs, emotional state, or neural data. The California Labor Federation supported similar bills in Massachusetts, Vermont, Connecticut, and Washington.
This article was originally published on CalMatters and was republished under the Creative Commons Attribution-NonCommercial-NoDerivatives license.
