In September, California Governor Gavin Newsom thought-about 38 AI-related payments, together with the extremely contentious SB 1047, which the state’s legislature despatched to his desk for last approval. He vetoed SB 1047 on Sunday, however signed greater than a dozen different AI payments into legislation over the course of the month. These payments attempt to tackle probably the most urgent points in synthetic intelligence: every little thing from futuristic AI methods creating existential danger, deepfake nudes from AI picture turbines, to Hollywood studios creating AI clones of useless performers.
“Residence to the vast majority of the world’s main AI corporations, California is working to harness these transformative applied sciences to assist tackle urgent challenges whereas learning the dangers they current,” mentioned Governor Newsom’s workplace in a press launch.
To this point, Governor Newsom has signed 18 AI payments into legislation, a few of that are America’s most far reaching legal guidelines on generative AI but. Right here’s what they do.
AI danger
On Sunday, Governor Newsom signed SB 896 into legislation, which requires California’s Workplace of Emergency Companies to carry out danger analyses on potential threats posed by generative AI. CalOES will work with frontier mannequin corporations, corresponding to OpenAI and Anthropic, to research AI’s potential threats to essential state infrastructure, in addition to threats that might result in mass casualty occasions.
Coaching information
One other legislation Newsom signed this month requires generative AI suppliers to disclose the info used to coach their AI methods in documentation revealed on their web site. AB 2013 goes into impact in 2026, and requires AI suppliers to publish: the sources of its datasets, an outline of how the info is used, the variety of information factors within the set, whether or not copyrighted or licensed information is included, the time interval the info was collected, amongst different requirements.
Privateness and AI methods
Newsom additionally signed AB 1008 on Sunday, which clarifies that California’s present privateness legal guidelines are prolonged to generative AI methods as properly. That signifies that if an AI system, like ChatGPT, exposes somebody’s private info (identify, tackle, biometric information), California’s present privateness legal guidelines will restrict how companies can use and revenue off of that information.
Schooling
Newsom signed AB 2876 this month, which requires California’s State Board of Schooling to contemplate “AI literacy” in its math, science, and historical past curriculum frameworks and educational supplies. This implies California’s colleges might start instructing college students the fundamentals of how synthetic intelligence works, in addition to the restrictions, impacts, and moral issues of utilizing the know-how.
One other new legislation, SB 1288, requires California superintendents to create working teams to discover how AI is being utilized in public faculty training.
Defining AI
This month, Newsom signed a invoice that establishes a uniform definition for synthetic intelligence in California legislation. AB 2885 states that synthetic intelligence is outlined as “an engineered or machine-based system that varies in its degree of autonomy and that may, for express or implicit aims, infer from the enter it receives the right way to generate outputs that may affect bodily or digital environments.”
Healthcare
One other invoice signed in September is AB 3030, which requires healthcare suppliers to reveal after they use generative AI to speak with a affected person, particularly when these messages comprise a affected person’s scientific info.
In the meantime, Newsom just lately signed SB 1120, which places limitations on how well being care service suppliers and well being insurers can automate their providers. The legislation ensures licensed physicians supervise the usage of AI instruments in these settings.
AI robocalls
Final Friday, Governor Newsom signed a invoice into legislation requiring robocalls to reveal whether or not they’ve use AI-generated voices. AB 2905 goals to stop one other occasion of the deepfake robocall resembling Joe Biden’s voice that confused many New Hampshire voters earlier this 12 months.
Deepfake pornography
On Sunday, Newsom signed AB 1831 into legislation, which expands the scope of present youngster pornography legal guidelines to incorporate matter that’s generated by AI methods.
Newsom signed two legal guidelines that tackle the creation and unfold of deepfake nudes final week. SB 926 criminalizes the act, making it unlawful to blackmail somebody with AI-generated nude pictures that resemble them.
SB 981, which additionally turned legislation on Thursday, requires social media platforms to ascertain channels for customers to report deepfake nudes that resemble them. The content material should then be briefly blocked whereas the platform investigates it, and completely eliminated if confirmed.
Watermarks
Additionally on Thursday, Newsom signed a invoice into legislation to assist the general public establish AI-generated content material. SB 942 requires broadly used generative AI methods to reveal they’re AI-generated of their content material’s provenance information. For instance, all pictures created by OpenAI’s Dall-E now want slightly tag of their metadata saying they’re AI generated.
Many AI corporations already do that, and there are a number of free instruments on the market that may assist individuals learn this provenance information and detect AI-generated content material.
Election deepfakes
Earlier this week, California’s governor signed three legal guidelines cracking down on AI deepfakes that might affect elections.
One among California’s new legal guidelines, AB 2655, requires giant on-line platforms, like Fb and X, to take away or label AI deepfakes associated to elections, in addition to create channels to report such content material. Candidates and elected officers can search injunctive reduction if a big on-line platform shouldn’t be complying with the act.
One other legislation, AB 2839, takes purpose at social media customers who submit, or repost, AI deepfakes that might deceive voters about upcoming elections. The legislation went into impact instantly on Tuesday, and Newsom urged Elon Musk could also be prone to violating it.
AI-generated political commercials now require outright disclosures below California’s new legislation, AB 2355. Which means transferring ahead, Trump might not have the ability to get away with posting AI deepfakes of Taylor Swift endorsing him on Reality Social (she endorsed Kamala Harris). The FCC has proposed an analogous disclosure requirement at a nationwide degree and has already made robocalls utilizing AI-generated voices unlawful.
Actors and AI
Two legal guidelines that Newsom signed earlier this month — which SAG-AFTRA, the nation’s largest movie and broadcast actors union, was pushing for — create new requirements for California’s media trade. AB 2602 requires studios to get hold of permission from an actor earlier than creating an AI-generated reproduction of their voice or likeness.
In the meantime, AB 1836 prohibits studios from creating digital replicas of deceased performers with out consent from their estates (e.g., legally cleared replicas have been used within the latest “Alien” and “Star Wars” motion pictures, in addition to in different movies).
SB 1047 will get vetoed
Governor Newsom nonetheless has a couple of AI-related payments to resolve on earlier than the top of September. Nevertheless, SB 1047 shouldn’t be one in all them – the invoice was vetoed on Sunday.
Throughout a chat with Salesforce CEO Marc Benioff earlier this month through the 2024 Dreamforce convention, Newsom might have tipped his hat about SB 1047, and the way he’s fascinated by regulating the AI trade extra broadly.
“There’s one invoice that’s kind of outsized when it comes to public discourse and consciousness; it’s this SB 1047,” mentioned Newsom onstage this month. “What are the demonstrable dangers in AI and what are the hypothetical dangers? I can’t remedy for every little thing. What can we remedy for? And in order that’s the strategy we’re taking throughout the spectrum on this.”
Test again on this text for updates on what AI legal guidelines California’s governor indicators, and what he doesn’t.