New York Starts RAISE Act Registration for Frontier AI

On September 21, 2026, Governor Kathy Hochul announced next steps to implement the Responsible AI Safety and Education Act, the RAISE Act. Her office says that starting in November, New York will direct large frontier AI developers to register with the state and prepare for compliance, and that beginning in January 2027 those companies must comply and report regularly to the Office of Digital Innovation, Governance, Integrity and Trust, known as DIGIT, which sits inside the Department of Financial Services. The governor appointed Marc Gilman, previously general counsel and vice president of compliance at Theta Lake, as Deputy Director for the RAISE Act, and her office describes him as DIGIT’s first full-time hire. She also said she will explore ways to build on the law in the coming months. The release closes by pointing to her July moratorium on new hyperscale data centers and to the Host Community Investment Framework that Empire State Development released the week before.

The law itself is Senate Bill S8828, the chapter amendment signed on March 27, 2026 as Chapter 96, which rewrote the original RAISE Act and set its effective date at January 1, 2027. It defines a frontier model as a foundation model trained using more than 10^26 integer or floating-point operations, counting fine-tuning and later material modifications, and a large frontier developer as one that, with its affiliates, had more than $500 million in annual gross revenue in the preceding calendar year. A frontier developer must report a critical safety incident to the DIGIT office within 72 hours, and within 24 hours to an appropriate authority when the incident poses an imminent risk of death or serious injury. Large frontier developers must publish a frontier AI framework, send summaries of catastrophic risk assessments from internal use every three months, and register, disclose and pay assessments. The attorney general can seek civil penalties of up to $1 million for a first violation and $3 million for each subsequent one. The scope clause covers frontier models developed, deployed or operating in whole or in part in New York.

Some trade coverage on September 21 described the announcement as a new round of transparency, safety and incident reporting requirements for data center developers and operators. PSV read the governor’s release and the full text of Chapter 96, and neither supports that framing. The release describes obligations on large frontier AI developers. The statute’s duties run to frontier developers and large frontier developers, and the phrase data center does not appear anywhere in it. The rules that do bind New York data center projects are still Executive Order 62, the July 14 pause on state environmental permits for new facilities of 50 megawatts or more, which this desk has covered. An owner or lender who files the RAISE Act under new data center regulation is filing it in the wrong place.

The indirect exposure is real, though, and it runs through the tenant. The scope clause turns on where a model is developed, deployed or operating, so a frontier developer’s decision to run a model from a New York data hall is one of the things that brings it under the law. The statute requires a large frontier developer’s framework to describe the cybersecurity practices that keep unreleased model weights from unauthorized modification or transfer by internal or external parties, and the first category of critical safety incident is unauthorized access to, modification of, or exfiltration of model weights that results in death or bodily injury. The catastrophic risk threshold includes more than $1 billion in damage to, or loss of, property from a single incident. PSV’s read, not something the law or the governor’s office addresses: where the weights sit in space a landlord secures, the landlord’s access control records, badge logs and incident facts become evidence the tenant may need, and the tenant needs them on a 72-hour clock. Expect security exhibits and notice clauses in leases with frontier developer tenants to tighten. For office owners the effect is smaller. The compute and revenue thresholds limit the law to a handful of the largest AI companies, not the typical AI tenant signing a floor in Manhattan.

The artifact is one row per lease, pre-lease or colocation agreement in New York data center or office space. Inputs: the lease, its security exhibit and its notice provisions; the tenant entity and its affiliates; the tenant’s stated use where the lease or marketing materials disclose it, training, inference or general cloud; any published frontier AI framework from the tenant or its parent; and the DIGIT registration list once the office publishes one. Output, per row: whether the tenant or its parent is plausibly a large frontier developer on the statute’s two tests, whether model weights or model operations are likely to sit in the space, the landlord’s current incident notice period and records retention for access logs, and a flag wherever the lease’s notice timeline is longer than the 72 hours the tenant owes the state.

The reviewer is the asset manager with leasing counsel, and the approval gate is simple: no security exhibit or incident notice clause for a frontier developer tenant is agreed until counsel has compared it against the tenant’s statutory reporting duty and the landlord’s actual ability to produce records in time. An assistant is well suited to the extraction, pulling notice periods, security obligations and records clauses out of lease files and flagging mismatches against the statute. It should not decide whether a tenant is covered by the law, which is the tenant’s determination and its counsel’s, and it should not characterize any incident.