New York did not abandon frontier-AI regulation, but the law now in force is materially different from the more interventionist bill lawmakers passed in 2025. The original RAISE Act emphasized safety plans, independent review and employee protections; the current framework centers on public disclosures, incident reporting and state oversight. A technology-and-academic coalition opposed the original bill, and universities appeared on its membership list—but that does not establish that each university approved the campaign or its advertising.
What was the original RAISE Act?
The Responsible AI Safety and Education Act, or RAISE Act, was introduced as S6953/A6453 and moved through amended versions, including S6953B/A6453B. Senator James Gounardes sponsored the Senate measure, with Assemblymember Alex Bores a principal Assembly sponsor. The Legislature passed the B versions in June 2025. The original proposal was aimed at frontier models and catastrophic risks, not ordinary consumer-facing AI errors. Its sponsor’s memo described harms on the scale of more than $1 billion in damage or hundreds of deaths or injuries. The Senate bill and sponsor’s memo set out the proposal’s rationale and provisions.
Its policy approach was comparatively direct: developers would maintain safety plans, submit them to qualified third-party review, protect employees who raised serious safety concerns, and disclose major security incidents. Those obligations would have required more than publishing general descriptions of a company’s safety practices.
How the law changed
“Defanged” is a judgment, not a term in the statute. It is defensible as a description of the shift from the Legislature’s original proposal, particularly the move away from a package built around safety plans, independent review and explicit employee protections. It would be misleading, however, to use the word to mean New York has no meaningful AI-safety requirements.
Recommended Free Tools
| Issue | Original legislative proposal | Current framework |
|---|---|---|
| Core approach | Safety plans, third-party review, employee protections and disclosure of major security incidents. Senate bill | Published frontier AI frameworks and model transparency reports, incident reporting and administrative oversight. Current chapter-amendment bill |
| Who is covered | Frontier-model developers under the definitions in the original proposal. Senate bill | A “large frontier developer” is a frontier developer whose affiliates collectively exceeded $500 million in annual gross revenue in the preceding calendar year; a frontier model is defined using a training-compute threshold above 1026 integer or floating-point operations, alongside statutory criteria. Current chapter-amendment bill |
| Risk focus | The sponsor memo framed the bill around catastrophic harms, including more than $1 billion in damage or hundreds of deaths or injuries. Senate bill | The current definition includes more than 50 deaths or serious injuries, or more than $1 billion in property damage, subject to statutory exclusions. Current chapter-amendment bill |
| Public information and review | Safety plans and third-party review were central features. Senate bill | Frameworks and model transparency reports must be published; third-party assessment is addressed within framework and disclosure obligations. The law allows specified redactions. Current chapter-amendment bill |
| Oversight | The original proposal had its own remedies and violations provisions. Senate bill | A revised regime overseen through an office within the Department of Financial Services. Current chapter-amendment bill |
The threshold figures make the law’s potential reach dependent on both a model’s training compute and the developer’s corporate structure and revenue. A company’s AI subsidiary might be smaller than the statutory revenue threshold on its own, while affiliated entities could push the group over it. The statute’s definitions, rather than a company’s prominence, determine coverage.
#1 Best Overall
Why there are two signing dates
Governor Kathy Hochul signed the RAISE Act on December 19, 2025, after negotiating amendments. Her announcement described requirements for large AI developers to publish information about safety protocols, report qualifying incidents to the state within 72 hours after determining that an incident occurred, and operate under oversight housed in the Department of Financial Services. The governor’s signing announcement describes that version and its reporting formulation.
That was not the last change to the statute. A subsequent chapter-amendment measure, A9449/S8828, was signed March 27, 2026. The legislative record says it repealed and replaced the earlier statutory article with a new Article 44-B. The current obligations discussed here are those in that later framework, not simply the December announcement. The A9449 record records the signing and current text; the S8828 record describes the companion measure.
Who campaigned against the original bill?
Reporting identified the AI Alliance, a coalition with technology-company and academic members, as an opponent of the original measure. The reported corporate members included Meta, IBM, Intel, Oracle, Snowflake, Uber, AMD, Databricks and Hugging Face. Listed university or academic members included New York University, Cornell, Dartmouth, Carnegie Mellon, Northeastern, Louisiana State, Notre Dame, Penn Engineering and Yale Engineering. The campaign coverage reports the membership list and the campaign’s activity.
Rank #2
That membership list is evidence of an institutional association with the Alliance; it is not, by itself, proof that every university’s leadership endorsed the bill opposition, authorized advertisements using its name, or agreed with every campaign claim. The same reporting said most listed universities did not respond to requests for comment. A membership, a formal institutional position, a logo in an advertisement and the views of individual researchers are distinct facts.
What the opposition campaign argued
According to the campaign coverage, advertisements began November 23, 2025, with the message “The RAISE Act will stifle job growth.” The campaign linked the bill to risks for New York’s technology economy, citing roughly 400,000 high-tech jobs and major investment. Those are campaign claims, not independently established measures of jobs that would have been lost because of the bill.
The same report put estimated ad spending at approximately $17,000–$25,000 and potential reach at more than two million people, based on Meta’s Ad Library. These are reported estimates, not audited totals or proof of actual audience exposure. The report on the campaign is the source for those figures.
Concerns about compliance costs, trade secrets, model definitions, open research and inconsistent state rules can be relevant to evaluating an AI law. But the available campaign account does not establish that a particular university joined the opposition for any one of those reasons, nor does it demonstrate that industry pressure alone produced the final text. The sequence—public opposition followed by negotiated amendments—is evidence of a political contest, not proof of who drafted each provision.
Free tools Windows power users keep installed
One-click scans. No signup required.
What the current law requires
The current Article 44-B framework applies to large frontier developers meeting the statutory definitions. It requires them to create, implement, follow and publish a frontier AI framework. The framework must address standards, thresholds for identifying catastrophic-risk capabilities, mitigations, review before deployment or extensive internal use, third-party assessment, cybersecurity for unreleased model weights, incident identification and response, governance, and risks arising from internal model use. It must be reviewed and, where appropriate, updated at least annually. Material modifications must be published with a justification within 30 days. The current statute’s text sets out these framework obligations.
Before or concurrently with deploying a new frontier model or a substantially modified version, a covered developer must publish a transparency report. Required information includes the release date, supported languages, output modalities, intended uses, generally applicable use restrictions, summaries and results of risk assessments, third-party evaluator involvement, and steps taken to comply with the framework.
Developers may redact information to protect trade secrets, cybersecurity, public safety, national security or compliance with other law. Where possible, they must describe the nature and justification of redactions, and retain unredacted information for five years. That balance matters: protections can guard sensitive details, but broad redactions can also make public disclosures less useful for independent scrutiny.
The law also requires reporting of qualifying critical safety incidents and creates oversight within the Department of Financial Services. Hochul’s December announcement stated a 72-hour reporting period after a developer determines an incident occurred; the operative requirements should be read in the current chapter-amendment text rather than inferred solely from the announcement.
Where the framework leaves difficult questions
- Coverage: The compute and revenue thresholds limit the law to a subset of developers. Whether a specific company or model is covered depends on the statutory definitions and affiliate calculations.
- Meaningful disclosure: A report can satisfy formal requirements while offering limited practical insight, particularly if assessments are summarized or sensitive details are redacted.
- Independent scrutiny: The statute includes third-party assessment in the framework, but the value of that review depends on evaluator independence and what findings become public.
- Risk boundaries: The catastrophic-risk definition excludes, among other things, information substantially similarly accessible from non-foundation-model sources, lawful federal-government activity, and harm where a model did not materially contribute when combined with other software.
- Changes to models: A substantially modified model may trigger a new transparency report. Fine-tuning or reinforcement learning could matter depending on whether the change meets the statutory standard.
- Internal use: The framework expressly addresses catastrophic risks from internal use, so not every consequential use is outside the law merely because the model was not released publicly.
These design choices show why “weakened” and “meaningless” are not synonyms. The current regime has real disclosure, governance, incident-reporting and oversight requirements, but it relies more on transparency and administrative review than on the original bill’s broader safety-plan and employee-protection architecture.
Best Value
What the university role does—and does not—show
Universities have plausible institutional interests on both sides of frontier-AI regulation. Partnerships with AI firms can bring research funding, cloud resources, models, internships and access for students; universities can also have legitimate concerns about whether rules are workable, whether disclosures compromise research or security, and whether state-by-state regulation creates inconsistent obligations. Researchers within one university may disagree about the right balance among transparency, liability, safety planning and open research.
Those are possible incentives and debates, not evidence that any named institution’s partnership caused it to oppose the bill. The reporting establishes that universities were listed among Alliance members and describes a campaign; it does not establish the extent of institutional authorization or internal agreement. That distinction is essential to an accountability story: scrutiny should ask who made decisions about membership and campaign use of institutional names, without turning association into proof of capture.
Was the bill defanged?
Compared with the proposal lawmakers passed, yes in a specific sense: the path to enactment removed or altered central elements of the original approach and left a framework more focused on public reporting, incident disclosure and agency oversight. But New York still has a frontier-AI law with substantive framework, assessment, cybersecurity, governance and transparency duties for developers that meet its thresholds. Whether those duties produce usable information and effective oversight will determine how much protection the statute delivers in practice.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
