Technology

California's governor vetoed a large AI safety bill

Law hammer

AI safety is one of the hottest topics in tech right now. Companies have scooped and scraped an untold amount of data (YOUR data) to train their AI tools, and many of those tools pose a threat to cybersecurity and job security. This is why governmental intervention is required to make sure that AI companies don’t go too far. However, California’s governor just vetoed a pretty important AI safety bill. Was it for a good reason? It’s tough to say.

What is this bill?

Many state politicians have undoubtedly been inundated with bill after bill to help regulate the burgeoning AI technology. California’s Governor, Gavin Newsom, mentioned that he has seen many, and signed off on them as well. However, one fell through the cracks. In a letter to the California State Senate, he explained that he decided to return bill SB 1047.

This bill outlines several key requirements that will affect AI companies. These companies would be required to implement strict safety protocols that would prevent unwanted bias, unintended outcomes, misinformation, etc. Next, these companies would be subject to regular audits to make sure that they’re complying with the safety protocols. The companies would also need to report any and all notable incidents to help the authorities identify emerging threats.

The bill would require that there be a cloud computing cluster created called a “CalCompute”. This would be a publicly available cluster of cloud computers that will help with safe AI development.

Lastly, it would need to ensure protection for whistleblowers who report on potentially harmful AI development or violations of the law. It will help ensure transparency.

Targeting the titans

SB 1047 would mostly target larger companies whose AI has the potential to affect people en masse. These include companies like Google, Meta, OpenAI, Microsoft, Anthropic, and others. Companies with models that cost over $100 million to train will be under this bill. Also, it would affect companies with models that cost $10 million to fine-tune. This is one important part of Newsom’s argument.

This bill seems to be in line with many of the regulations proposed over the past few years. It pushes for transparency, government oversight, and restrictions for major AI companies.

California’s governor vetoed an important AI safety bill

As useful as this bill sounds, Gavin Newsom had other plans. He explained why he didn’t want to pass the bill. He picked apart some of the key aspects of the bill.

SB 1047 takes the approach that many regulatory bodies have taken; targeting the largest companies with the most potential to cause harm. This approach makes sense. However, in his statement, he mentioned that there’s always the possibility that the biggest threat might not be the biggest company. “Smaller, specialized models may emerge as equally or even more dangerous than the models targeted by SB 104,” he said in his letter.

Rather than only focusing on the larger companies, he wants to have regulations that will scale to affect smaller companies as well. The small-time chatbot that keeps you company could pose a risk along with Google’s massive umbrella of products.

Too broad

Another point that Newsom brought was the fact that SB 1047 is rather broad. He says that it “does not take into account whether an AI system is deployed in high-risk environments, involves critical decision-making or the use of sensitive data. Instead, the bill applies stringent standards to even the most basic functions – so long as a large system deploys it.

Newsome doesn’t think that this approach will work. The bill doesn’t really take into account what the AI is being developed for; it will need to be restricted so long as it’s being made by a large company. So, it doesn’t matter if a tool is being made to regulate the flow of power in a city, track the progress of a pandemic, or just help you write witty social media posts; if a large company develops it, then it will need to be regulated.

What Newsom wants are regulations that focus on what areas the AI can do the most damage, not what companies have the most capital.

Some people aren’t happy

As you can guess, there are some people who aren’t too happy about this decision. Scott Wiener, the main author of the bill, spoke out about the veto on X.com. “This veto leaves us with the troubling reality that companies aiming to create an extremely powerful technology face no binding restrictions from the U.S. policymakers,” he said in his statement. He finished by saying “we’re not going anywhere.” So, it’s obvious that we’ve not heard the last from him or other voices advocating for regulation.

In fact, other notable supporters of the bill include Elon Musk, Mark Hamill, Alyssa Milano, Shonda Rhimes, and J.J. Abrams. Not only that, but the bill also got support from SAG-AFTRA, SEIU, and other unions. There are some pretty heavy figures looking to have this bill go through.

Who opposed the bill? This should come as no surprise

There are several who are happy about the veto, and it should come as no surprise that many of them are companies developing AI. Companies like Google, Meta, and OpenAI were vocal about their opposition to the bill. Fewer guardrails mean more wiggle room for them to develop their AI. This is for better or worse.

For example, a Google Spokesperson told the Verge, “We join many small and larger developers in thanking Governor Newsom for helping California continue to lead in building responsible AI tools that benefit California, our nation, and the world.

Over on Meta’s side, the company’s public affairs manager told The Verge, “We are pleased that Governor Newsom vetoed SB1047. This bill would have stifled AI innovation, hurt business growth and job creation, and broken the state’s long tradition of fostering open-source development. We support responsible AI regulations and remain committed to partnering with lawmakers to promote better approaches.”

On a separate note, OpenAI’s chief strategy officer, Jason Kwon, wrote a letter to Scott Wiener telling him that the bill would stifle progress.

These are the typical responses that we’d expect from the companies developing AI technology. It’s true that strict regulation would make it tougher for them to progress and keep up with competition.

Other people who opposed the bill include former House Speaker Nancy Pelosi, San Francisco Mayor London Breed, and eight congressional Democrats from California.

Was this the right move?

At its surface, this news seems like it would be frustrating for people pushing for AI regulation. Right now, the AI regulatory framework is like a patchwork of potential bills, hopes of laws, and major fragmentation. A state-wide tightening of the vice for AI companies in California would have been huge for the AI industry. As Newsom puts it, 36 of the 50 biggest AI companies are in California. This means that California holds a ton of power in the AI landscape.

There needs to be some solid regulation for these companies. Let’s not kid ourselves, we’ve seen how defiant companies like Meta and Google are when it comes to staying within the law. How many times have these companies been smacked on the wrist for violating data privacy laws? We don’t know the full effects of AI on the world just yet, so the companies with the largest sway over the AI industry should be the ones with the tightest muzzles.

The Bill aims to address that. It would have taken away their ability to self-police their actions. So, we’re left without any strong legislature that would have otherwise kept these companies on the straight and narrow.

But…

We need to look at this veto from both sides. Major companies need to be regulated. However, we can’t doubt that the bill has some holes in it that Newsom pointed out. Should companies be restricted when developing tools for small tasks? If Google develops a tool that creates AI-generated avatars, should the company have to report everything on that?

Also, there’s also the point that he made about the bill only targeting large companies. Again, we still don’t know the full scope of AI’s effects. So, it’s hard to tell if an up-and-coming company could create a tool that could have some devastating effects. We just don’t know. A lion can cause a ton of damage, but so can a mouse.

We should take into account that the bill has its blind spots. So, it’s hard to say whether the veto was the right move, the wrong move, or something in between. What we can hope for is a new bill that will be closer to everyone’s liking.