The SB-1047 Bill and the question of AI regulation at the US State level

Written by Bernat Baella i Masuet

California is known for being the home of technological companies like Apple, Google and Meta. Due to a mix of universities, research centers, entrepreneurial spirit, and human talent, the state is a hub of innovation. Artificial Intelligence (henceforth, AI) is no exception to this.

As more companies develop their AIs, regulations try to keep pace. California has been one of the first US states to do so. The state’s legislative process follows a series of steps, after which, the proposed bills get sent to the Governor, who can sign it, allow the bill to become a law without signing it or veto it. Only in the last case does the bill not become a law.

In the last months, several AI bills have been signed into law in California, including AB-3030, which requires healthcare providers to disclose when they use Generative AI to communicate with patients, AB-2905, which makes robocalls inform that they are using AI-generated voices, or AB-1831, that includes content generated by AI systems as child pornography. As seen with the examples, the laws do not focus on a single sector or activity but rather try to be as comprehensive as possible.

Among the bills sent to be signed by the Governor of California, one has stood out among the rest: SB-1047 Safe and Secure Innovation for Frontier Artificial Intelligence Models Act. The AI community in California is not a homogenous bloc. On the one hand, Open AI, and groups representing Meta, owner of Facebook, and Google, opposed it on the grounds that it would be bad for AI development. On the other hand, Elon Musk, one of the founders of Open AI who later left the company due to disagreements with the direction of the company, and AI scientists such as Geoffrey Hinton and Yoshua Bengio defended the need for the regulation provided by the bill. They cited the risks presented with AI models and the need for companies to implement safeguards. Apparently, the arguments presented by the opposing parties were more compelling, since the Governor of California, Gavin Newsom, vetoed the law, preventing it from entering into force.

Among the measures proposed by the bill, the most controversial ones were making companies liable for harms made by AI models and forcing companies to introduce a “Kill switch” for AI systems.

The implicit threat made by the opposing parties to this bill is that they would consider relocating somewhere else where these restrictions were not in place. Jason Kwon, OpenAI’s Chief Strategy Officer, warned about this in a letter sent to Senator Scott Wiener, one of the authors of the SB-1047 Bill. This situation poses a conundrum for California’s regulators: to regulate or not to regulate, or rather, up to which point to regulate, to avoid overstepping the mark.

If they put very strict requirements and safeguards, some companies might choose to move to other US states where the regulation about AI is looser or nonexistent, with the laissez-faire approach applying to a greater degree. Since the companies would still be present in the US, the state could do little to prevent them from doing so. It is also true that few states offer the advantages of California, or that the enterprises might not end up carrying out the threat, with few or no relevant companies moving out. Some relevant companies have left the state in the past, with the expansive regulation of the state being cited as one of the reasons. Examples include Fortune 1000 companies like McKesso, a pharmaceutical distributor, oil companies like Chevron, or technological ones such as Tesla or Oracle. Hence, even though not all companies would leave, some of the relevant ones could, affecting the state’s position as a “tech hub” and frontrunner in innovation.

The other alternative is not regulating AI or watering down the legislation, making compliance easier. This, in theory, would prevent an exodus of companies to other states, preserving the role played until now by California. This option, however, does not seem to be preferred by the state’s legislators, since they have previously passed several pioneering regulations.

However, a fact remains. California’s Senate or Assembly are not the European Commission, meaning that they do not regulate a territory as big as the EU. This obvious fact, however, is not always taken into account when crafting legislation. If the EU enacts legislation, companies have a much bigger incentive to comply. They can threaten to leave, but then they lose access to 16% of world GDP and a mature market. If companies were to leave California, they would need to find new offices and relocate workers, but apart from that, which is more logistical than economical, there would not be many more relevant effects.

Thus, the basic question remains: up to what point can California legislate without companies leaving the state in search of kinder ones? The veto of this particular bill alone does not mean much, but seeing the trend followed by companies that did end up leaving the state and the federal situation, in which no relevant laws comparable to the ones in California have been passed, might indicate that the state can’t do much else to regulate AI comprehensively. That is, without risking an exodus of leading companies. This view may be shared by Gavin Newsom, who despite pressure from legislators, might decide to wait for the new President to enact federal legislation, eliminating the incentive for companies to relocate within the US.

Only time will tell what happens. Perhaps no new laws similar to the SB-1047 Bill will be passed in California, perhaps an amended or watered-down version will pass, or maybe the new US President might decide to regulate it federally. The lesson is that regulating a matter such as AI is not as straightforward as in other areas, since companies might relocate relatively easily to avoid the strictest legislation. Thus, a need for international cooperation to regulate it is highly needed to avoid fragmentation and a cherry-picking of countries by enterprises based on how loosely they regulate AI.

Sources:

Overview of Legislative Process. (2019). Ca.gov. http://www.leginfo.ca.gov/bil2lawx.html T. Howell, & B. R. Allen. (2024). Decoding California’s Recent Flurry of AI Laws [Review of Decoding California’s Recent Flurry of AI Laws]. https://www.foley.com/insights/publications/2024/10/decoding-california-recent-ai-laws/ Allyn, B. (2024, September 29). California Gov. Newsom vetoes AI safety bill that divided Silicon Valley. NPR. https://www.npr.org/2024/09/20/nx-s1-5119792/newsom-ai-bill-california-sb1047-tech Rhim, R. (2024, April 23). The Corporate Departure from California. SCBC; SCBC. https://www.scbc-law.org/post/the-corporate-departure-from-california

Eurostat. (2020). China, US and EU are the largest economies in the world. https://ec.europa.eu/eurostat/documents/portlet_file_entry/2995521/2-19052020-BP-EN.pdf/ bb14f7f9-fc26-8aa1-60d4-7c2b509dda8e