社團指南

New York RAISE Act for Frontier AI Models

New York enacted the RAISE Act in 2025, then amended and replaced its original statutory text in 2026; the current framework takes effect January 1, 2027.

  • 4 分鐘閱讀
  • 最後更新
本頁4 分鐘閱讀
  1. 概述
  2. 深入探討
  3. 戰略影響
  4. The Future of New York RAISE Act for Frontier AI Models
  5. 現實世界的實施
  6. 風險與防護欄
  7. 實施路線圖
  8. 不斷探索
  9. 常見問題

概述

It sets transparency and incident-reporting duties for frontier developers and added duties for large frontier developers, defined in part by more than $500 million in annual gross revenue. Critical safety incidents generally must be reported within 72 hours, with a separate 24-hour disclosure for imminent death or serious physical injury risks.

深入探討

New York first enacted the RAISE Act in 2025. Governor Hochul signed Chapter Amendment S8828 on March 27, 2026; it replaced Article 44-B and set its effective date to January 1, 2027. The amended text controls. It defines a frontier model as a foundation model trained using more than 10^26 integer or floating-point operations, counting original training and subsequent fine-tuning, reinforcement learning, or other material modifications. A frontier developer trains or initiates training at that threshold. A large frontier developer and its affiliates exceeded $500 million in annual gross revenue in the preceding calendar year. Large frontier developers must publish and follow a frontier AI framework covering catastrophic-risk thresholds and assessment, mitigations, third-party assessment, deployment review, cybersecurity for unreleased model weights, incident response, internal governance, and risks from internal use. They review the framework at least annually and publish a justification within 30 days after a material modification. Before deployment of a new or substantially modified frontier model, frontier developers publish a transparency report; large frontier developers include summaries of assessments and mitigation steps. All frontier developers have incident-reporting duties under the amended text. A critical safety incident must be reported to the state office within 72 hours after a determination or after learning facts sufficient for a reasonable belief that one occurred. If the developer discovers that the incident poses an imminent risk of death or serious physical injury, disclosure to an appropriate authority is required within 24 hours. Specified violations carry penalties up to $1 million initially and $3 million for repeats; no private right of action is created. The 2026 amendment changed the Act’s scope and effective date. A compliance inventory should separate frontier developers from large frontier developers, identify affiliates and compute, and track the office’s implementing rules. The statute applies to frontier models developed, deployed, or operating in whole or part in New York, subject to exceptions and any designated federal-equivalence pathway.

戰略影響

風險與安全

災難性和日常的人工智慧危害都取決於誰了解風險以及誰能夠採取行動。

更明確的決策

民眾和專業素養決定強而有力的安全政策在政治上是否可行。

突破炒作

清晰的解釋可以減少炒作、實驗室公關和模糊道德劇場的影響。

The Future of New York RAISE Act for Frontier AI Models

The amended Act becomes effective January 1, 2027, leaving time for agencies and developers to prepare. The Department of Financial Services office may issue rules and administer assessments, and a federal-equivalence pathway may affect incident reporting. Recheck the current chapter and regulations before deployment; do not rely on summaries of the superseded 2025 text. Watch for implementation rules from the New York Department of Financial Services office and any federal-equivalence designation. Update thresholds and procedures before the 2027 effective date.

現實世界的實施

A covered large frontier developer publishes its frontier AI framework and summarizes catastrophic-risk assessments before deploying a new or substantially modified model.

A frontier developer that determines a critical safety incident occurred reports it to the state office within 72 hours; an imminent death or serious-injury risk also triggers a 24-hour disclosure to an appropriate authority.

A developer evaluates whether its model exceeds the current statutory compute threshold and whether affiliate revenue makes it a large frontier developer.

A smaller frontier developer below the large-developer revenue threshold distinguishes baseline transparency and reporting duties from large-developer framework and filing duties.

風險與防護欄

  • 將存在風險視為科幻小說,同時能力複合。

  • 混淆了表面產品安全與高度自治下的對準。

  • 只給非英語和非專業觀眾留下低品質的資源。

實施路線圖

  1. 單獨的產品危害、誤用和失控/失調風險。

  2. 詢問哪些證據會改變您對時間表和嚴重性的看法。

  3. 比起行銷主張,更喜歡主要來源和具體評估。

  4. 確定一條行動路徑:職業、政策、資金或技能——而不僅僅是意識。

不斷探索

Free newsletter

Get the daily AI briefing

Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.

One email each weekday. Unsubscribe in one click. We never sell or share your address.

Test yourself

Take the New York RAISE Act for Frontier AI Models quiz

Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.

開始測驗

Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation

常見問題

What is New York RAISE Act for Frontier AI Models?

New York enacted the RAISE Act in 2025, then amended and replaced its original statutory text in 2026; the current framework takes effect January 1, 2027. It sets transparency and incident-reporting duties for frontier developers and added duties for large frontier developers, defined in part by more than $500 million in annual gross revenue. Critical safety incidents generally must be reported within 72 hours, with a separate 24-hour disclosure for imminent death or serious physical injury risks.

When does the current amended RAISE Act framework take effect?

The 2026 chapter amendment sets January 1, 2027 as the effective date for the replacement Article 44-B.

What compute threshold defines a frontier model under the amended text?

The statute defines frontier models using more than 10^26 integer or floating-point operations, including qualifying later modifications.

How is a large frontier developer distinguished from another frontier developer?

The definition uses preceding-calendar-year annual gross revenue above $500 million, counting affiliates.

Which deadline applies to an ordinary critical safety incident under the amended Act?

The amended statute sets a 72-hour period from determination or learning facts sufficient for reasonable belief.

What additional clock applies when an incident poses an imminent risk of death or serious injury?

The law requires disclosure to an appropriate authority within 24 hours for that imminent risk.