California Considers Adoption of SB1047, Safe and Secure Innovation for Frontier Artificial Intelligence Models Act

California is on the verge of adopting a controversial bill that would impose unprecedented new regulations on the development of AI: SB 1047: Safe and Secure Innovation for Frontier Artificial Intelligence Models Act.  A full copy of the bill has been linked here.  If adopted, the proposed bill would establish unprecedented new compliance obligations for AI software start-ups in Silicon Valley.

What is SB 1047, the Safe and Secure Innovation for Frontier Artificial Intelligence Models Act?

The Safe and Secure Innovation for Frontier Artificial Intelligence Models Act (the “AI Models Act”) would create a new Frontier Model Division within California’s Department of Technology which would have oversight powers over the training of many new AI models.  Under the AI Models Act, developers of these AI models would be required to build a so-called kill switch into the AI model and to potentially shut down the model until the Frontier Model Division deems that the AI model is subject to a “limited duty exemption,” which would be defined as:

a determination. . . . that a developer can provide reasonable assurance that the covered model does not have a hazardous capability, as defined, and will not come close to possessing a hazardous capability when accounting for a reasonable margin for safety and the possibility of posttraining modifications.

A “covered model” under the AI Models Act would be defined to mean an “artificial intelligence model that was trained using a quantity of computing power greater than 10^26 integer or floating-point operations, and the cost of that quantity of computing power would exceed one hundred million dollars ($100,000,000) if calculating using average market prices of cloud compute as reasonably assessed by the developer at the time of training.

As currently proposed, “derivative” AI models would be exempt from the new compliance obligations: only “non-derivative” AI models would be subject to the obligations.

A “derivative model” would be defined to be an artificial intelligence model that is derivative of another AI model, including either ” a modified or unmodified copy of an artificial intelligence model” or “a combination of an artificial intelligence model with another software.  The “derivative model” would be specifically defined not to include “an entirely independently trained artificial intelligence model” or an “artificial intelligence model, including one combined with other software, that is fine-tuned using a quantity of computing power greater than 25 percent of the quantity of computing power, measured in integer or floating-point operations, used to train the original model.”

What constitutes a “hazardous capability” under the proposed legislation?

The AI Models Act would define “hazardous capability” to constitute the capability of a covered model to be used in one of the following harms:

• the creation or use of a chemical, biological, radiological, or nuclear weapon in a manner that results in mass casualties
• at least $500 million dollars of damage through cyberattacks on critical infrastructure via a single incident or multiple related incidnts
• at least $500 million dollars of damage by an AI that autonomously engages in conduct that would violate the Penal Code if taken by a human
• bodily harm to another human
• the theft of or harm to property
• other grave threats to public safety and security that are of comparable severity to the harms described above.

Penalties for noncompliance with this legislation would include punitive damages and a civil penalty for a first violation not to exceed ten percent of “the cost of the quantity of computing power used to train the covered model to be calculated using average market prices of cloud compute at the time of training” and 30 percent of the same in case of a second violation.  The legislation authorizes joint and several liability against the developers directly where

(1) steps were taken in the development of the corporate structure among affiliated entities to purposely and unreasonably limit or avoid liability.

(2) The corporate structure of the developer or affiliated entities would frustrate recovery of penalties or injunctive relief under this section.

What has been the reaction to SB 1047 from the Silicon Valley start-up community?

Bloomberg also reported that the a key point of contention in the startup community is the idea that AI developers are responsible for people who misuse their systems, pointing to Section 230 of the Communications Decency Act of 1996, which has shielded social media companies from liability over content users create on platforms.

Author Jess Miers of the Chamber of Progress criticized the legislation on the basis that it would “introduce a high degree of legal uncertainty for developers of new models, making the risks associated with launching new AI technologies prohibitively high.”

The Silicon Valley Software Law Blog will continue following legislative developments relating to SB 1047 as this bill advances.

If you have questions regarding your software company’s potential compliance obligations under SB1047, please schedule a consultation with me at this link.