The RAISE Act has overwhelmingly passed the New York Assembly (95-0 among Democrats and 24-22 among Republicans) and New York Senate (37-1 among Democrats, 21-0 among Republicans).
It's a shame that whistleblower protections were cut from the RAISE Act in its last iteration.
In version A (June 3) they were still included, providing protections for a wide range of individuals in flagging critical concerns internally or the AG directly. Arguably written in a better way than e.g. proposed in the recently introduced AI Whistleblower Protection Act.
Hi OAISIS! Bill sponsor here. We did not give that up lightly. However, I think there were two things that contributed to cutting that:
1. We think that if we pass the rest of the bill, then most (if not all) of the conduct that was covered in the whistleblower section is already covered in the New York Labor Law. Article 740 says employees cannot be retaliated against if an employee:
"(a) discloses, or threatens to disclose to a supervisor or to a public
body an activity, policy or practice of the employer that the employee
reasonably believes is in violation of law, rule or regulation or that
the employee reasonably believes poses a substantial and specific danger
to the public health or safety;
"(b) provides information to, or testifies before, any public body
conducting an investigation, hearing or inquiry into any such activity,
policy or practice by such employer; or
"(c) objects to, or refuses to participate in any such activity, policy
or practice."
2. While NY Labor Law is very expansive in terms of the behavior it protects, it's punitive fines are puny, only $10,000 per employee. There was some concern that if we were repeating that number in the RAISE Act, it may be used as a model in other states that have higher baseline penalties.
Hi Alex - very much appreciate the response and context. We had suspected as much on Argument 1, Argument 2 was new to us and is appreciated.
My impression on the 'loss' here is that RAISE would've introduced protections for non-paid advisors and gone further on contractor protections - if we understand it correctly, e.g. Eval Providers would've then be covered, which they are not under 740.
And RAISE's text around 'substantial' vs. 'substantial and specific' danger in 740.
Having said that: It's still great to see this pass and having violations of RAISE be covered under Section 740 protections now is a much better choice than to NOT have RAISE pass at all.
What exactly is the attorney general going to do with a "security incident" where he's completely in over his head to understand the implications? There's a handful of people who understand the correct course of action and none of them work for the government.
If you believe in slowing AI progress in general this is maybe a good thing, assuming it'll give alignment tech time to catch up? At this point I unironically think a frontier model is more aligned with human values than any national government, and them being unregulated is a feature, not a bug. The alignment work can only happen at AI labs, the rest of humanity is sitting on the sidelines regardless of how upset it makes them.
I believe it applies to anyone who makes their models available in New York, so the California-based labs would have to comply or disable access within the state. Complying would likely be much cheaper.
Podcast episode for this post:
https://open.substack.com/pub/dwatvpodcast/p/rtfb-the-raise-act
It's a shame that whistleblower protections were cut from the RAISE Act in its last iteration.
In version A (June 3) they were still included, providing protections for a wide range of individuals in flagging critical concerns internally or the AG directly. Arguably written in a better way than e.g. proposed in the recently introduced AI Whistleblower Protection Act.
In version B (June 9) they were cut completely.
Hi OAISIS! Bill sponsor here. We did not give that up lightly. However, I think there were two things that contributed to cutting that:
1. We think that if we pass the rest of the bill, then most (if not all) of the conduct that was covered in the whistleblower section is already covered in the New York Labor Law. Article 740 says employees cannot be retaliated against if an employee:
"(a) discloses, or threatens to disclose to a supervisor or to a public
body an activity, policy or practice of the employer that the employee
reasonably believes is in violation of law, rule or regulation or that
the employee reasonably believes poses a substantial and specific danger
to the public health or safety;
"(b) provides information to, or testifies before, any public body
conducting an investigation, hearing or inquiry into any such activity,
policy or practice by such employer; or
"(c) objects to, or refuses to participate in any such activity, policy
or practice."
2. While NY Labor Law is very expansive in terms of the behavior it protects, it's punitive fines are puny, only $10,000 per employee. There was some concern that if we were repeating that number in the RAISE Act, it may be used as a model in other states that have higher baseline penalties.
Separately, Zvi, thank you for this write-up!
Hi Alex - very much appreciate the response and context. We had suspected as much on Argument 1, Argument 2 was new to us and is appreciated.
My impression on the 'loss' here is that RAISE would've introduced protections for non-paid advisors and gone further on contractor protections - if we understand it correctly, e.g. Eval Providers would've then be covered, which they are not under 740.
And RAISE's text around 'substantial' vs. 'substantial and specific' danger in 740.
Having said that: It's still great to see this pass and having violations of RAISE be covered under Section 740 protections now is a much better choice than to NOT have RAISE pass at all.
Thank you. And some things for us to work on next year!
hehe pubic hearing.
Thanks for the post, it's a note of hope
What exactly is the attorney general going to do with a "security incident" where he's completely in over his head to understand the implications? There's a handful of people who understand the correct course of action and none of them work for the government.
If you believe in slowing AI progress in general this is maybe a good thing, assuming it'll give alignment tech time to catch up? At this point I unironically think a frontier model is more aligned with human values than any national government, and them being unregulated is a feature, not a bug. The alignment work can only happen at AI labs, the rest of humanity is sitting on the sidelines regardless of how upset it makes them.
What I think confusing is… who cares ? It’s not binding for California-based labs, is it ? So it doesn’t apply to anyone ?
I believe it applies to anyone who makes their models available in New York, so the California-based labs would have to comply or disable access within the state. Complying would likely be much cheaper.
This would be preempted by the current version of the federal budget bill, right?