Over the course of September 2024, Californian Governor Gavin Newsom signed multiple bills into law relating to generative AI. Signature, or at least approval, by the Governor is the final step in the Californian legislative process. The increase in regulation relating to GenAI follows a 2023 executive order which already included multiple provisions on Generative AI, including state employee training, legislative engagement, and a risk-analysis report. The recent bills cover a wide range of topics and sectors, further illustrated below:

Performers’ likeliness - Two bills, AB 2602 and AB 1836, were driven by SAG-AFTRA, the performers’ union addressing AI-related concerns within the entertainment industry, in response to concerns about AI’s ability to digitally replicate voices and likenesses without proper consent or representation. AB 2602 states that contract provisions allowing for the use of digital replicas of a person’s voice or likeness for work in place of the person themselves, or use of their voice or likeness to train GenAI system, are only valid if the provisions clearly define and detail all uses of the digital replica or AI system, and if the performer was represented by legal counsel or a labour union. Meanwhile, AB 1836 protects the rights of deceased performers, and their heirs, by requiring permission from the performer’s heirs or estate before use of digital replicas of their likeness or voice in an expressive audiovisual work or sound recording.

AI watermarking - SB 942 requires widely used generative AI systems to include provenance disclosures (i.e. data verifying the authenticity or origin) in the content they generate which can then be detected by free AI detection tools that should be offered by the provider together with the system.

State use of Gen AI - SB 896 requires government bodies to further assess the risk of using generative AI on critical state infrastructure and annually report to the legislature. Additionally, the bill requires state agencies and departments that use GenAI when directly communicating with a person regarding government services to disclose the use of the AI system and inform persons of how to contact a human employee.

Deepfake election content – AB 2655 requires large online platforms to remove or label deceptive deepfakes related to elections during specific periods before an election, as well as provide platform users with a way to report such content. AB 2355 requires political campaigns to disclose the use of generative AI in advertisements.

GenAI transparency - On 26 and 28 August 2024, the California legislative chambers passed, respectively, AI bills AB 2013 and SB 1047. AB 2013, mandates transparency about the data used to train generative AI systems, including through a high-level summary, and has since been signed into law by the governor. SB 1047 on the other hand would have required developers of advanced AI models (i.e. models trained using a certain amount of computing power) to integrate significant safety measures to mitigate potential risks posed by AI, such as the capability for a full shutdown and use of a third-party auditor. However, the bill has been vetoed by the governor who reasoned that regulation should apply to AI systems based on the threats they pose instead of based on their computing power during training. Finally, the governor also signed AB 3030, which requires specific health care providers to disclose the use of GenAI when used in communications to patients pertaining clinical information.

Protection of children - Protections for children are a key focus as well, with AB 1831 and SB 1381 expanding child pornography statutes to include AI–generated and AI–altered content. SB 976 also tries to protect children from social media addiction by requiring social media apps to not provide addictive content feeds to users under 18 without parental consent. Notifications from addictive internet-based services or applications are also limited during school hours and between midnight and 6 am.

Sexually explicit deepfakes – SB 926 criminalises AI-generated sexually explicit deepfake content that would cause serious emotional distress to the person depicted in the deepfake. Additionally, under SB 981, social media platforms must also create a mechanism for users to report sexually explicit deepfakes of themselves, causing the content to be blocked.

Inactive proposals - While many bills made it through the legislature, others, such as AB 2930 and AB 3211, failed to pass and have been marked as inactive. AB 2930 addressed discriminatory AI. It would have required an impact assessment performed by AI deployers and developers, submitted to the Civil Rights Department. It also would have made unlawful discriminatory AI illegal and subject to a fine of $25,000 for each violation. The other bill that didn’t make it to the governor’s office is AB 3211, which would have required all generative AI developers to automatically add provenance data to synthetic content as well as have required platforms to fulfill several transparency obligations relating to such provenance data.

California is not the only state in the USA taking action on AI regulation. In 2024, several states have introduced legislation aimed at governing AI. California's recent developments in this area could encourage other states to follow suit, similar to trends seen with other tech-related issues. Increased activity at the state level may also prompt Congress to explore federal AI legislation more seriously.