Published on December 10, 2025 6:47 PM GMT
Summary
If you live in New York, you can contact the governor to help the RAISE Act pass without being amended to parity with SB 53. Contact methods are listed at the end of the post.
What is the RAISE Act?
Previous discussion of the bill:
- RTFB: The RAISE Act. This dives into the bill in detail, which I mostly omit from this post.
- Consider donating to Alex Bores, author of the RAISE Act: this post doesn’t explain the substance of the bill more than the RTFB, but it does add some co…
Published on December 10, 2025 6:47 PM GMT
Summary
If you live in New York, you can contact the governor to help the RAISE Act pass without being amended to parity with SB 53. Contact methods are listed at the end of the post.
What is the RAISE Act?
Previous discussion of the bill:
- RTFB: The RAISE Act. This dives into the bill in detail, which I mostly omit from this post.
- Consider donating to Alex Bores, author of the RAISE Act: this post doesn’t explain the substance of the bill more than the RTFB, but it does add some color around the passage of the bill.
You can read the bill yourself: S6953B. It is dry legalese, but it is short, with the PDF clocking in around 4 pages.
If you really don’t have time, a one sentence summary: the bill requires leading AI labs to explain their safety standards, and notify the government when one of their AIs does something bad, with the death of 100 people being definitely bad.
I assume that we’re on the same page that this is excessively basic AI regulation that should have been passed years ago.
We have SB 53, why would we want the RAISE Act?
The RAISE Act has several notable differences from California’s similar SB 53 which seem better for AI safety:
- It plainly states “a Large Developer shall not deploy a Frontier Model if doing so would create an unreasonable risk of Critical Harm.” SB 53 only penalizes violations of the developer’s own framework[1].
- It focuses criteria on the compute costs going into models, where SB 53 also takes into account revenue. This can cover an AI lab like Safe Superintelligence Inc. which does not plan to generate revenue in the near future, but could plausibly train a frontier model[2].
- Keep in mind that if SSI never actually deploys their hypothetical frontier model, the RAISE Act doesn’t actually come into effect.
- It contains provisions for distillations, so large distillations of Frontier Models are definitely still considered Frontier Models.
- It explicitly calls out IP transfers as transferring Large Developer status, so a Large Developer can’t dodge these reporting requirements by training a model under one corporate entity, and then transferring the model to another entity.
- This does not mean individual users of open weight/open source models suddenly become Large Developers, requiring “a person subsequently transfers full intellectual property rights of the frontier model to another person… and retains none of those rights for themself…”
- Reporting timelines are (mostly) shorter, 72 hours (3 days) versus 15 days in SB 53.
- The 72 hours matches cybersecurity reporting timelines (example NYS bill).
- SB 53 does contain an extra provision where “imminent risk of death or serious physical injury” requires an additional shorter disclosure timeline of 24 hours, with looser disclosure to “an authority” instead of the usual Office of Emergency Services.
- Larger penalties, with a maximum $30M penalty, whereas SB 53 has a $1M maximum.
- The safety incident definition in RAISE Act is slightly broader, for example including “a frontier model autonomously engaging in behavior other than at the request of a user” providing “demonstrable evidence of an increased risk of critical harm”, where the closest comparable SB 53 clause requires the model to be deceptive.
These are small differences (along with other differences not highlighted here), but they seem worth having versus not.
Unfortunately, the RAISE Act is already neutered from an earlier draft.
I accidentally read the earlier A draft of the bill first, which was much more aggressive (and therefore interesting!) than the final draft we got. Interesting provisions that draft A contained which were removed:
- A requirement for a 3rd party audit. SB 53 also does not require 3rd party evaluation, although it does require describing any such evaluations.
- A requirement to prepare a Safety and Security Protocol in pre-training, if the developer expected to be a Large Developer by the end of the training run.
- Clauses to pierce the corporate veil, making it harder to dodge liability. This sounds cool to my layperson ears, but it’s unclear to me whether this is as aggressive and unusual as it sounds.
- Detailed broad whistleblower protections, along with $10k per employee penalties for retaliation. SB 53 defines special whistleblower protections just for safety personnel, but it also references a broad California whistleblower law (1102.5), which covers all employees and contains a retaliation civil penalty of $10k per employee. This is less exciting, since it seems like SB 53+1102.5 already covers this angle.
Look at what they took from us! If the RAISE Act was substantially draft A (with the various editorial mistakes smoothed out), making sure it passed and wasn’t changed to be more like SB 53 would have been a much clearer win.
The RAISE Act is not strictly better for safety than SB 53.
Some notable differences:
- RAISE has a “critical harm” threshold at 100 casualties, where SB 53 has a tighter “catastrophic risk” threshold at 50 casualties.
- SB 53 contains a preemption clause, overriding local laws and regulations. The RAISE Act does not contain a parallel clause, which might open the door to, say, New York City trying to override the RAISE Act.
- I am not 100% certain that RAISE will actually apply to anything? SB 53 has the advantage that all the large AI labs like OpenAI, Anthropic, and Google are headquartered in California, so SB 53 definitely applies to them. Meanwhile RAISE needs to include “Scope. This article shall only apply to frontier models that are developed, deployed, or operating in whole or in part in New York State.” Does this include New York residents going to chatgpt.com and typing into it, even if the ChatGPT servers are outside NYS? Does this conflict with the commerce clause? From my layman perspective I would guess that the law would end up intact after challenges, but boy oh boy I am not a lawyer!
The RAISE Act may be substantially modified
The RAISE Act has passed the New York State legislature in June, and seems likely to signed by the governor.
Prediction markets put passage at 80%.
- A Manifold market with moderate liquidity has 80% as of December 8th. Note that the market will still resolve to “Has become law” even if the bill is substantially weakened.
- Another Manifold market with a simpler structure and somewhat less liquidity has 80% as of December 8th.
- Kalshi, Polymarket, and Predictit do not have relevant markets when searching for “raise act”.
However, the governor can negotiate for modifications in a NYS specific process called chapter amendments. Kathy Hochul, the current governor, has used chapter amendments more than any other governor, amending 1 out of 7 bills (or 1 out of 3?).
Does Hochul want to significantly amend the RAISE Act? As best as I can tell, this is highly uncertain. Data points that seem relevant:
- In August, Hochul mentioned “seeking a balance” on the RAISE Act (paywall).
- In September, Hochul mentioned “it’s hard when one state has a set of rules, another state does” but also right afterwards “a lot of companies should be adopting these internal controls themselves”.
- In October, Alex Bores, one of the bill sponsors, threw an AI centered fund-raiser on Hochul’s behalf, which raised $250k (NYT paywall).
- The same NYT article cites a tech industry association president requesting Hochul to change the bill to be closer to SB 53.
- On December 2nd, Alex Bores (again, a sponsor of the RAISE Act) appeared in a video and said he expects chapter amendment conversations “in the next few weeks”.
- Procedural actions hint that Hochul expects the bill to be controversial (see foldout below).
- There’s a fair amount of spend on ads and lobbying on opposing the RAISE Act. On the other hand, the populace at large seems to support RAISE (see foldout below).
As such, I have no idea what probability I would put on whether the RAISE Act would be significantly amended, or how we would even define “significantly amended”. Uncertainty!
Details and implications of the NYS legislative process
The RAISE Act was passed in June, at the end of the 2025 NYS legislative session. The governor needs to sign a bill within 10 days or it will be automatically vetoed, but a tacit administrative understanding extends this timeline: the legislature will not send a bill to the governor unless the governor requests it, and the 10 day deadline doesn’t start until the bill is sent to the governor.
In this way the governor can continue signing bills until the very end of the year (or beyond? One source claims the governor also has January). Usual practice seems to be for the governor to leave dealing with controversial bills to the end of the year, either to allow the most information to trickle in or to use the holidays as cover.
As of midday December 9th the governor has 161 bills on her desk, with only 5 bills not yet requested (via doing NYS bill searches, 3 when searching for “Passed Senate” and 2 for “Passed Assembly”), including the RAISE Act.
As of the end of day on December 9th, the RAISE Act is finally marked as delivered to the governor.
The late delivery implies that the bill is controversial. This could be good or bad! Hochul might be set on signing the bill, and plans on using the holidays as cover to sign it without Andreessen Horowitz making a huge ruckus. Or, Hochul might be planning on negotiating down to parity with SB 53, and is leaving the bill to the end of the year to get those concessions. Someone more familiar with NYS politics might be able to read this better, but it is fairly unclear to me.
For his part, Alex Bores (RAISE Act Assemblyman sponsor) expects chapter amendments without sounding too worried about them, but maybe it’s because it’s poor political form to appear panicked; unfortunately, a calm, cool, collected appearance isn’t much evidence for good or bad outcomes.
Sources: general explainer by Reinvent Albany, RAISE specific explainer.
It seems difficult to get a better base rate for the chances the RAISE Act will be substantially changed in chapter amendments.
There are a few things that we could look into:
- Do bills signed in December have a higher rate of chapter amendments? As discussed in the previous foldout, common wisdom holds that more controversial bills are passed in December, so they may have a meaningfully different rate of chapter amendments.
- Chapter amendments may be innocuous or substantial; what is the base rate of substantial chapter amendments?
Unfortunately these questions are not easy to answer.
The rate of chapter amendments for December bills is more straightforward, it just requires a ton of leg work. Using the NY Assembly bill search, we can easily find bills passed (chaptered?) in December of 2024, which returns 268 bills. However, there isn’t structured data exposing just chapter amendments, so now we would be stuck doing 268 searches for bills that might have chapter amendments.
Determining whether chapter amendments are substantial is subjective, and requires reading the full text of all the bills involved, which is a lot of work. It is possible to automate this with LLMs (if you trust them!), but one would still need to gather all the bill text and find applicable chapter amendment relationships.
It is also not clear if Hochul’s rate of chapter amendments is 1 out of 7 from New York Focus in Jan 2024, or 1 out of 3 from Alex Bores (RAISE sponsor) in Dec 2025. Even our basest base rate is uncertain! We could resolve this with the same sort of chapter amendment analysis we talked about before, but applied to all the bills in the past few years.
I would estimate that it would take around a week of work to answer these questions, but it’s unclear to me that it would be helpful. Surely someone would have noticed if 90% of December bills had chapter amendments, or 90% of chapter amendments were substantial changes under Hochul, so it seems unlikely that the differences are that large. I estimate that an incredibly small number of people would be sensitive to smaller (but still significant) differences: that is, if our hypothetical leg work found that the rate of December bill chapter amendments was 100% over the base rate (14% (1/7) to 28%), I would expect that almost everyone that previously would not have called the governor would still not call the governor.
What does the landscape of support for/opposition to the RAISE Act look like?
I suspect that there is a fair bit of unobserved lobbying happening. As an example, the Leading the Future PAC is spending large sums on defeating RAISE Act sponsor Alex Bores’ current congressional bid, specifically citing the RAISE Act. I lack the resources to find out if the backers of the PAC are also lobbying Hochul to veto or modify the RAISE Act, but come on, it’s probably happening.
We also see opinion columns and reports coming out against the RAISE Act, with bursts of activity in June and October (looking through X/Twitter posts mentioning “RAISE Act”, and following links to the underlying articles):
- Statement by the Chamber of Progress (2025-06-12). Also interesting to see Alex Bores and Daniel Eth show up in the tweet replies.
- Opinion column (2025-06-05).
- Probably an ad campaign; how else do you have a 30s unlisted video with 4.7M views?
- Times Union opinion column (2025-07-24) (Times Union is a real newspaper). Interestingly, the author Marc Alessi (a publc figure!) also wrote another opinion column a few days later, on 2025-07-31.
- NY Daily News opinion column (2025-10-08) (paywalled). Daily News is a real newspaper.
- Opinion column #1 and opinion column #2 are exactly the same article, with the same page layout but with different brand names, published on the exact same day. Is this a content farm?
- Syracuse.com opinion column (2025-10-31). Syracuse.com is attached to a real newspaper.
- Crain’s opinion column (2025-10-22) (paywalled). Crain’s New York Business is a real paper.
- Computer & Communications Industry Association 2025 AI Landscape report (2025-11-19).
One interesting thing is that many of the tweets linking to these columns/statements have zero engagement. Putting on my cynical cap, someone could be trying to buy a change in public opinion, but is just doing it incredibly poorly.
Unfortunately, the arguments invariably push willful misreadings of the bill, raising questions about whether the authors have read the bill at all. If only we had worthy opponents!
What about support for the RAISE Act?
- A poll by Beacon Research showed 84% support for RAISE among New Yorkers (2025-06-09). The site is a bit janky, but maybe it’s just relentlessly mobile focused?
- @Eko_Movement took out a full page ad.
- @DanceNYC called for contacting Hochul (2025-11-13). I didn’t expect the dancers to be on board, but hell yeah, happy to have them.
- @STOPSpyingNY held events and called for contacting Hochul (2025-12-03).
- There’s obvious AI Safety adjacent support, which I won’t list here.
The thing is that, at least on X/Twitter, there isn’t obviously much more engagement on support vs opposition posts. Unfortunately, I didn’t see the STOP Spying NYC event until it was already over, which would have been one way to gauge in person enthusiasm. I hoped this little aside would provide evidence for whether a handful of voters contacting the governor would be impactful, but I genuinely have no idea.
Is the RAISE Act about to be preempted?
So far federal legislative preemption efforts have stalled, but there’s reports that the federal executive may issue an execuctive order preempting state level AI regulation, which would include the RAISE Act. However, Trump hasn’t actually signed the EO yet (as of December 9th), and it might not even be upheld: if it was legal to use an EO to preempt AI regulations, why did they spend all that effort trying to do the preemption through legislation first? The bill might be put into legal limbo even before it is signed, but it does seem better to have it in abeyance (and ready to go if preemption is cleared out of the way) instead of needing to pass a new law later down the line.
What can be done?
In the end I am incredibly uncertain about basic facts like where the governor stands, or how entrenched she is in her position. However, contacting the governor is incredibly cheap (a few minutes of time)[3], so it’s hardly punishing to be wrong.
I would expect the greatest impact to come from New York residents contacting the governor. In order of expected impact:
- Call the governor, Kathy Hochul: 518-474-8390 (9 AM to 5 PM, Monday to Friday).
- Fill out the “Send a Message to the Governor” form, which I would guess is equivalent to email.
- Use the PassRaiseAct.com form letter.
If you only have time to fill out the form letter, then do that. If you have more time, you could customize your message to focus specifically on keeping the bill as-is, or supporting the additional provisions of the RAISE act over SB 53, which the form letter doesn’t include. Additionally, my impression is that customized messages are weighted more than form letters, so calling or emailing without following a script (or using a script you’ve personalized) is better.
Snail mailing a letter is an option as well, but given the tight timelines it might not arrive on time. My received folk wisdom for political feedback effectiveness is that letters are better than calling which is better than email, but I have no idea whether modern political back offices consider the method of contact when aggregating information on constituent concerns.
The RAISE Act was delivered to the governor on December 9th. The governor has 10 days (excluding Sundays) to sign the bill or it will be pocket vetoed, so by my reckoning the latest day the bill can be signed is December 20th. I’m uncertain when Hochul will sign it, so if you plan on contacting the governor you should do it sooner than later[4].
What is calling the governor like?
When I called the governor’s line on December 9th, the phone tree offered me two options:
- Leave a message offering ideas or opinions to the governor.
- Connect with an agent.
I decided to leave a message, roughly following:
“Hello! I am $NAME, calling from $AREA (making it clear that I’m a resident of NYS), calling to register support for the RAISE Act, namely the provisions that go above and beyond SB 53. As an example, I think that it is important to keep $PROVISION in the RAISE Act. Thank you for your time.”
I did stumble over my words a bit more than this would imply, but I’m sure the staff listening to the messages are used to it. It is unclear whether all calls get this treatment, or if calling at times (like 4pm) will be moved to a phone tree. That is to say, I’m not sure if it’s possible to be jump scared by a live agent without interacting with the phone tree first.
I’m not sure if out of state residents calling will be helpful, given my understanding is that out of state calls are seen as a threat or signal that the caller will make future political donations, but there was already a fairly successful fundraiser (NYT paywall), so maybe that signal is less impactful now. On the other hand, I can’t see how out of state calls would hurt.
h/t Zvi’s RTFB; I missed the actual prohibition on risk for many drafts.
h/t Transformer News, I missed the revenue implication for many drafts.
By my reckoning I spent 20 hours on researching and writing this over the last week, so it wasn’t especially cheap for me, but hopefully I’ve made it cheap for you!
One possibility is that Hochul is racing Trump’s preemption EO, which seems good for the RAISE Act passing intact.
Discuss