-2

The latest proposed version of "SB-1047 Safe and Secure Innovation for Frontier Artificial Intelligence Models Act" can be found here. Since it regulates a service (artificial intelligence models) rather than physical products, its hard to figure out what its jurisdiction is supposed to be. Does it cover only AI models built in California? Only models sold in California? Every model in the world?

The law is mute on jurisdictional limits but presumably some must exist, as California has ~zero power to impose anything outside its own borders.

JonathanReez
  • 4,816
  • 4
  • 37
  • 70

2 Answers2

4

It doesn't regulate a service. It defines certain requirements for developers and providers of certain computational capabilities:

22603 (a) Before beginning to initially train a covered model, the developer shall do all of the following: ...

22604 (a) A person that operates a computing cluster shall implement written policies and procedures to do all of the following when a customer utilizes compute resources that would be sufficient to train a covered model: ...

And imposes civil penalties for violations:

22606 (a) The Attorney General may bring a civil action for a violation of this chapter and to recover all of the following: ...

See this answer that explains how, unless explicitly said otherwise, the jurisdiction is assumed to be territorial. I.e.: individuals and companies in California are to be covered by the law.

ohwilleke
  • 257,510
  • 16
  • 506
  • 896
littleadv
  • 8,641
  • 2
  • 17
  • 45
2

The law is mute on jurisdictional limits but presumably some must exist, as California has ~zero power to impose anything outside its own borders.

That sounds like a non-sequitur to me, it makes very little sense to limit the scope of a law based on that. If anything, it seems existing statutes (not only in the US) do the exact opposite and either

  • Claim very broad application (e.g. universal jurisdiction, jurisdiction based on the location or citizenship of a consumer or victim)
  • Deliberately limit their scope where to avoid conflicts of norms or serve various interests (e.g. anti-bribery rules that explicitly do not apply to foreign operations of local businesses)

The former is useful to affirm a principle or make a political or diplomatic point even if it has limited practical impact. The latter is useful to constrain law enforcement and the justice system, precisely because enforcement would be possible without this limitation. By contrast, you don't need to spell out what is or is not materially impossible in law, that's not what norms are for.

In a comment, you mention an hypothetical model “released in Arizona” but it's not directly relevant to SB-1047 and somewhat unclear what it would mean in practice. You seem to be wishing for some definition that could be exploited to create a loophole but it's hard to see why that would be good for California. From the perspective of the drafters of the law, it's much better to create some uncertainty to nudge as many businesses as possible to focus on implementing the law instead of circumventing it.

In practice, just as US financial sanctions are pretty hard to ignore, it will be difficult to run a (tech) business with zero connection to California, be it the infrastructure used, employees or business locations, or customers, especially if you consider the thresholds to trigger the rule (only very costly models are covered, which imply vast ressources and large investments that are hard to imagine without Silicon Valley VC-money and customers). So if it chooses to leverage any of these to pressure businesses, California definitely has a lot of power to influence what's being done beyond its borders.

Relaxed
  • 1,174
  • 6
  • 6