The New York Public Service Commission has opened Case 26-M-0552 and ordered twelve named electric, gas and water utilities to file, within 60 days, a written inventory of every AI use case in their operations and the policies, procedures and protocols that govern them

Two filings, no permission slip

The order does two things, and page 2 states both. It orders the Subject Utilities to inventory and report all use cases of AI systems in their operations, and it directs them to disclose the policies, procedures and protocols that govern the use of AI systems within their organizations. Ordering clauses 3 and 4, on pages 9 and 10, turn those into filing duties, each due within 60 days of issuance.

That is the whole operative content. The order does not require Commission approval before a utility deploys an AI system, does not restrict any deployment, and does not direct any utility to build a testing regime or an override capability it does not already have. Clause 4 asks for the documents that exist. Where the agency looks further ahead, it does so in the press release rather than in the ordering clauses, and we take that up below.

Ordering clause 6 reads, in full, that this proceeding is continued. The order sets no second reporting date and no recurring cadence. We searched the ten pages for any semiannual or periodic reporting requirement and found none.

Twelve named companies, filing one by one

Page 2 names the Subject Utilities and clauses 2, 3 and 4 repeat the list: Central Hudson Gas & Electric Corporation; Consolidated Edison Company of New York, Inc.; Liberty Utilities (St. Lawrence Gas) Corp.; Liberty Utilities (New York Water) Corp.; New York State Electric & Gas Corporation; Niagara Mohawk Power Corporation d/b/a National Grid; The Brooklyn Union Gas Company d/b/a National Grid NY; KeySpan Gas East Corporation d/b/a National Grid; National Fuel Gas Distribution Corporation; Orange and Rockland Utilities, Inc.; Rochester Gas and Electric Corporation; and Veolia Water New York, Inc.

Clauses 3 and 4 say each shall individually submit. Three of the twelve trade as National Grid and two as Liberty Utilities, and nothing in the order says a group parent may file once for its subsidiaries. Two of the twelve are water companies, which is why the legal authority section on page 7 cites PSL section 89-c alongside section 66.

PSEG Long Island is treated differently. Page 2 says the Commission encourages it to participate and provide the same information, for efficiency and consistency of the statewide review. It does not appear in any ordering clause. A headline saying every New York utility has been ordered to file is wrong on the order's own text.

What the order counts as an AI system

Page 2 defines an AI system as a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments. Footnote 1 sources that wording to 15 U.S.C. section 9401(3) and to New York State Information Technology Policy NYS-P24-001. The definition then names machine learning, large language models, natural language processing and computer vision, including generative AI, as included but not exhaustive.

The exclusion in the same footnote is where scoping arguments will start. The definition does not include basic calculations, basic automation, or pre-recorded rule-based conditional logic response systems with predefined triggers that automatically initiate predetermined actions. An outage prediction model that makes predictions against human-defined objectives and does not rest on the excluded basic calculations or predetermined rule-based responses is inside; the page 3 reference to outage prediction model development is an example of AI use, not a ruling that every such model qualifies. A protective relay that trips a predetermined action on a fixed threshold is, on the footnote's words, outside. Much of what sits between those two will need a judgement call, and the order supplies no further test.

What goes in each filing, and the 60-day clock

Page 8 sets the content of the inventory. The reports shall be detailed and descriptive regarding the AI system used, the functional objective of the system, and the level of human oversight. Clause 3 adds that the report is written, covers all use cases of AI systems in the utility's operations, and is to be consistent with the body of the order.

The governance filing carries the longer list. Pages 8 and 9 say the disclosure shall include, but not be limited to, AI Governance policies, verification methodologies, pre-deployment testing, human oversight protocols, override capabilities, and data privacy practices. Clause 4 makes that a filing of the policies, procedures and protocols themselves.

Both clauses run 60 days from the issuance of the order. The order was issued 17 September 2026, and 60 calendar days from that date is 16 November 2026. That date is our arithmetic; the order prints no calendar deadline. Clause 5 also matters. In the Secretary's sole discretion the deadlines may be extended, on a written request that includes a justification and is filed at least three days before the affected deadline.

Footnote 22 on page 8 anticipates that some of the material may fall within Public Officers Law sections 87 and 89. It tells each utility to evaluate whether a well-founded basis exists to file such information with the Department's Records Access Officer under POL section 89(5), and notes that section 89(5) envisions a substantive and tailored written supporting statement. That is an invitation to make a specific case for specific material, not a blanket confidentiality route.

The review standard is the Public Service Law, not a named framework

The last sentence before the ordering clauses, on page 9, states the test: the Commission will review these governance procedures and protocols in light of the utilities' obligations under the PSL. Those obligations are the ones the legal authority section on page 7 sets out, including the power under PSL section 66(5) to examine supervised corporations and keep informed as to their methods, practices, regulations and property.

NIST and the OECD appear on page 6, in the background section. The order describes the NIST AI Risk Management Framework and the OECD principles and observes that most AI governance frameworks share organizational accountability, transparency, explainability and human oversight as core principles. It does not adopt either document, does not order compliance with either, and does not name either in the ordering clauses.

The same-day press release, reference 26084/26-M-0552, goes one step past the order. It says the Commission will evaluate the procedures and protocols against commonly accepted AI Governance frameworks. That is the agency describing its intended approach in a release; the order's own formulation is the PSL. A utility drafting its filing should expect its documents to be read with both in mind, but only one of the two is in the instrument.

The background section also carries a worked example on pages 5 and 6: a 2025 Southern California Edison wildfire model that, the order says, was apparently misconfigured to stop forecasting damage after the eight-hour mark, citing Reuters. The order presents this as reported context for a risk category, not as a finding about any New York utility, and neither do we.

What we did not verify

What we opened: the full ten-page Order Initiating Proceeding, retrieved from the DPS Document and Matter Management system as item 1 in Case 26-M-0552 (DocRefId A044B0A0-0000-CD33-BEE0-07C138F2F784) and read end to end, including footnotes 1 to 22 and ordering clauses 1 to 6; the two-page press release filed as item 2 in the same case (reference 26084/26-M-0552); and the case's document list, read through its JSON endpoint because the DMM case page itself is JavaScript-rendered.

What we did not open: the Public Service Law sections cited in footnotes 16 to 21, 15 U.S.C. section 9401(3), NYS-P24-001, the NIST AI RMF, the OECD principles, the Public Officers Law provisions in footnote 22, the press and research sources in footnotes 2 to 15, and any utility filing in the case, of which the document list showed none at the time of reading. We describe the wildfire example and the framework discussion only as the order describes them.

What we refuse to claim: we do not say the order imposes semiannual or any recurring reporting, because it does not. We do not say it mandates compliance with the NIST AI RMF, the OECD principles or any named framework, because it reviews against the PSL and adopts none. We do not say PSEG Long Island or any utility outside the twelve is ordered to file. We do not say the order bans, licenses or requires approval of any AI deployment, or requires new testing or override controls, because it asks for existing policies and an inventory. We do not say any New York utility has caused harm through AI, because the order makes no such finding. We do not give a count of use cases or an estimate of filing burden, because the order gives none. 16 November 2026 is our calculation, not a printed date, and it is subject to clause 5. Where we use shall or must, we are quoting or closely reporting ordering clauses that are in force.

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.

Related briefings

• New York LOADING Act on state-agency automated decision systems
• NIST AI RMF critical infrastructure profile
• New York Executive Order 62 on hyperscale data centers
• North America AI regulation tracker
• AI Regulation News hub