California AI Transparency Act Passes Senate Judiciary Committee

[Update September 19, 2024: Governor Newsom signed SB 942 into law. Check out this blog post for a full analysis of SB 942, aka the California AI Transparency Act.]

I drove to Sacramento on April 16, 2024, and testified to the California Senate Judiciary Committee on behalf of State Senator Becker’s California AI Transparency Act (aka Senate Bill 942 or SB 942) The bill passed the Committee 10-1, with one Republican voting yes on it (making it bipartisan in support!). SB 942 now moves on to the California Senate Committee on Governmental Organization, where there will be a hearing on April 23, 2024. I was privileged to have proposed some of the ideas behind the bill — many of which leverage U.S. Senators Schatz and Kennedy’s proposed AI Labeling Act — and helped draft it, so it was a thrill to be able to testify on its behalf. The Senate Judiciary Staff did a thorough bill analysis, which I will summarize below with direct quotes.

Executive Summary of the Bill

This bill places obligations on businesses that provide generative artificial intelligence (AI) systems to develop and make accessible tools to detect whether specified content was generated by those systems. These “covered providers” are required to include visible and imperceptible markings on AI-generated content to identify it as such. The bill requires providers to register with the California Department of Technology (CDT).

What problems does the bill solve?

Certain forms of media – audio recordings, video recordings, and still images – can be powerful evidence of the truth. While such media have always been susceptible to some degree of manipulation, fakes were relatively easy to detect. The rapid advancement of AI technology, specifically the wide-scale introduction of generative AI models, has made it drastically cheaper and easier to produce synthetic content created, audio, images, text, and video recordings that are not real, but that are so realistic that they are virtually impossible to distinguish from authentic content, including so-called “deepfakes.” This bill works to ensure that providers of these generative-AI systems are equipping consumers with a tool to identify when specific content has been generated by their systems.

Nitty-gritty details of the bill

1) Establishes the California AI Transparency Act.

2) Requires a covered provider to create an AI detection tool by which a person can query the covered provider as to the extent to which text, image, video, audio, or multimedia content was created, in whole or in part, by a generative AI system provided by the covered provider that meets all of the following criteria:

a) The AI detection tool shall be publicly accessible and available via a uniform resource locator (URL) on the covered provider’s internet website and through its mobile application, as applicable.

b) The AI detection tool shall allow a person to upload content or a URL.

c) The AI detection tool shall support an application programming interface (API) that allows a person to invoke the AI detection tool without visiting the covered provider’s website.

d) The AI detection tool shall allow a person to provide feedback if the person believes the AI detection tool is not properly identifying content that was created by the provider.

3) Prohibits a covered provider from doing any of the following in carrying out the duties above:

a) Reveal personal information that identifies who utilized the AI system to create AI-generated content that was submitted to the AI detection tool.

b) Collect and retain personal information when a person utilizes the covered provider’s AI detection tool, except that it may collect and retain the contact information of a person who submitted feedback.

c) Retain any content submitted to the AI detection tool for longer than is necessary to comply with this law.

4) Requires a covered provider to include in AI-generated image, text, video, or multimedia content created by its own generative AI system a visible disclosure that meets all of the following criteria:

a) The disclosure shall include a clear and conspicuous notice, as appropriate for the medium of the content, that identifies the content as generated by AI, such that the disclosure is not avoidable, is understandable to a reasonable person, and is not contradicted, mitigated by, or inconsistent with anything else in the communication.

b) The disclosure shall, to the extent technically feasible, be permanent or difficult to remove.

c) The output’s metadata information shall include an identification of the content as being generated by AI, the identity of the tool used to create the content, and the date and time the content was created.

5) Requires a covered provider to include in AI-generated image, audio, video, or multimedia content created by its generative AI system an imperceptible disclosure that is machine detectable and is, to the extent technically feasible, permanent or difficult to remove.

6) Requires a covered provider to implement reasonable procedures to prevent downstream use of a generative AI system it provides without the disclosure required by this section, including the following:

a) Contractually requiring third-party licensees of the generative AI system to refrain from removing a required disclosure.

b) Terminating access to the generative AI system when the covered provider has reason to believe that a third-party licensee has removed a required disclosure.

7) Requires CDT, at least once every two years, to review these provisions and make recommendations to the Legislature regarding any amendments needed to account for changing technology and standards.

8) Requires covered providers to register with CDT and provide a URL to any AI detection tool it has created. Requires CDT to create and display on its website a Generative AI Registry that displays the names of registered covered providers and a link to the covered provider’s AI detection tool. CDT is authorized to charge a registry fee to covered providers to cover administrative costs.

9) Authorizes CDT to adopt regulations necessary to perform its duties hereunder.

10) Provides that a covered provider that violates the above provisions is liable for a civil penalty in the amount of $5,000 per violation to be collected in a civil action filed only by the Attorney General. Each day that a covered provider is in violation shall be deemed a discrete violation.

11) Defines the relevant terms, including:

a) “Artificial intelligence” or “AI” means a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments by using machine-based inputs and human-based inputs to perceive real and virtual environments, abstract its perceptions into models through analysis in an automated manner, and use model inference to formulate options for information or action.

b) “AI-generated content” means any form of digital content that is created with deep learning or machine learning processes.

c) “Covered provider” is a business that provides a generative AI system that has, on average over the preceding 12 months, over 1,000,000 monthly visitors or users and is publicly accessible within the geographic boundaries of the state.

d) “Generative AI system” refers to deep learning models that can generate text, images, and other content based on the data they were trained on.

Assuming the bill passes the next Committee, will do more blogging on it to explain some of its key points.

And the other bit of good news was I got to enjoy a gorgeous day in Sacramento.