California lawmakers push AI regulation bill amid industry backlash

California lawmakers are moving forward with another attempt to regulate artificial intelligence, over the objections of technology groups that say the latest bill is unworkable and possibly illegal. AB 412 seeks to thread the needle of demanding disclosure from companies that operate large language models, without directly restricting their behavior.

"It is no secret that AI developers use copyrighted materials to train their models," said the bill's author, Assemblymember Rebecca Bauer-Kahan, D-Orinda, presenting the bill to the Assembly Privacy and Consumer Protection Committee. "Just last week, OpenAI and Google came out publicly to say they needed to train on copyrighted materials to continue the work they are doing. When developers use copyrighted materials that they download from the Internet to train AI, they are not obtaining consent from the right owner of those materials and the owner often has no way of knowing."

An environmental attorney and chair of the committee where she presented on Tuesday afternoon, Bauer-Kahan has taken up the role as the most prominent watchdog of California's high-tech industry.

Her latest bill advanced on a 9-2, mostly party-line vote after she agreed to take several amendments. These include language specifying that the rules apply specifically to generative AI models. They also state that developers only must make a "reasonable effort" to identity copyrighted materials and developers who make all their data public are exempted.

"This bill doesn't change copyright law," Bauer-Kahan said. "We do not have the power to do that here in the state. What it does is it merely provides the copyright holders with notice of when their work has been put into an AI data set that trains an AI model."

She added that it would be up to copyright holders to vindicate their rights through lawsuits, discovery, and other standard methods of enforcing copyrights.

The bill is backed by several organizations representing creative professionals in Hollywood and elsewhere. Joely Fisher, secretary treasurer of SAG-AFTRA, spoke at the hearing in favor of the bill. She spoke about how her late half-sister, the actor Carrie Fisher, appeared in "Star Wars" films after her death with help of AI--but also with "permission and compensation."

"My union has no desire to halt technological progress," Fisher said.

She added, "AI can't do anything on its own and no AI algorithm is able to make something out of nothing. So, if intellectual property and copyrighted materials are being used to train AI models, the copyright owners need to know."

But the amendments still do not address the problems technology companies will face in trying to comply with AB 412, said Megan Stokes, the state policy director at the Computer & Communications Industry Association. Reached after the hearing, she said that it is difficult to determine what works are copyrighted online. The proportion of the works online that are copyrighted is very small, she said, and there are immense amounts of copyrighted material already being shared even before AI came on the scene.

"I got a little nervous when the sponsor was saying, 'Well the tech companies will figure this out, they're very smart people,'" Stokes said. "It's a little tough."

She added, "I appreciate the fact that the sponsor is looking at different amendments. I just really worry the bill is still not going to be technically workable."

Stokes also noted that AI developers are already working to comply with a law signed last year, AB 2013. That law states that beginning Jan. 1, 2026, AI models provide documentation of the data sets used to create them.

The hearing came on the same day the Joint California Policy Working Group on AI Frontier Models released its initial draft report. Gov. Gavin Newsom created it in September to create a policy framework for AI.

The report discussed at length the need to balance the benefits and risks of AI, while setting up reporting systems and protecting whistleblowers who identify abuses. It also argued that AB 2013 was a step toward catching up with the regulation model being assembled in the European Union.

"California passed and the governor signed AB 2013 in the 2024 legislative session, which requires public-facing transparency into the training data used to build generative AI systems," the report read. "The EU AI Act requires similar public-facing transparency into training data and copyright policies along with further disclosures on a range of topics, including risk mitigation."