A bill to regulate artificial intelligence companies in California has cleared a critical hurdle in the California State Assembly’s Budget Committee, but it’s still far from certain to pass. The California State Senate passed the bill earlier this year by a vote of 32 to 1, and now the bill will go before the full State Assembly for a vote. The date for that vote has not yet been set, but the State Assembly session ends on August 31, leaving the bill with a small window of time to pass the legislature.

If the bill is passed by the House, it will go to California Governor Gavin Newsom (D-CA), who must decide by the end of September whether to sign or veto the bill. Newsom has not commented on the bill, but has previously said that if it becomes “over-regulated,” California could “cede its latitude to other states or countries.”

The proposal by California State Senator Scott Wiener (D) would require safety testing for the most advanced AI models. This threshold is defined as models that cost more than $100 million to develop or that have a certain level of computing power. These safety tests would be required before the models could be made public. In addition, the California State Attorney General would have the power to bring charges against companies if their technologies are found to have caused serious harm.

The legislation has faced opposition from many technology companies, including some of the biggest tech giants like Google, Meta, Anthropic and OpenAI, as well as venture capital funds like A16z and Y Combinator, as the New York Times And TechCrunch. The concern of these corporations is that the legislation, while only a potential state law, could become a de facto national standard if other states decide to copy the law, or it simply becomes easier to comply with the law nationally than setting specific rules for users in California. Critics have also argued that the legislation could stifle innovation or push AI development out of California if the regulations are too burdensome, primarily due to the liability that companies would have to bear under the requirements.

Some of the attacks on the bill came particularly from California Democrats in Congress, including former House Speaker Nancy Pelosi (D-Calif.). Pelosi called the bill “well-intentioned but ill-informed” in a statement. She spoke in favor of introducing safeguards and guardrails for AI, but did not make any concrete proposals herself. Other California Democrats who criticized the bill include Representatives Zoe Lofgren (D-Calif.) and Ro Khanna (D-Calif.), among others. They called on Newsom to block the bill if it passes.

The draft bill, passed by the state Assembly’s budget committee, made industry-friendly changes to the bill, including removing a provision that would have established a Frontier Model Division in state government to accredit AI developers, eliminating civil penalties for violations that do not cause harm or risk, and adjusting language to require companies to exercise “reasonable care” to prevent harm from their AI systems rather than provide “reasonable assurances.” Enforcement authority remains with the attorney general.

Many of these changes reflect suggestions made by Anthropology Committee Chair Buffy Wicks in a letter published in July. The most significant change appears to be the adjustment of language from “reasonable assurance” to “reasonable care.”

Wiener has accepted the changes, but it is uncertain whether these changes will go far enough to appease opposition from AI companies, investors and lawmakers. In the coming days, as the bill is introduced to the full California State Assembly, more commentary is likely that will show how the chances of the bill passing in the House have changed after these changes. Newsom’s silence on the issue suggests that if frustration remains but is not enough to stop the bill in the State Assembly, he may be willing to be persuaded to veto the measure, especially if the opposition can convince him that it would still do more harm than good for the AI ​​industry.