Int. No.
By The Speaker (Council Member Menin)
A Local Law to amend the administrative code of the city of New York, in relation to third-party validation and shut-down capability of artificial intelligence models
Be it enacted by the Council as follows:
Section 1. Chapter 5 of title 20 of the administrative code of the city of New York is amended by adding a new subchapter 27 to read as follows:
SUBCHAPTER 27
ARTIFICIAL INTELLIGENCE MODELS
§ 20-883 Definitions. As used in this subchapter, the following terms have the following meanings:
Artificial intelligence model. The term “artificial intelligence model” means an engineered or machine-based system that varies in its level of autonomy and that can, for explicit or implicit objectives, infer from the input it receives how to generate outputs that can influence physical or virtual environments.
Behavior under distribution shift. The term “behavior under distribution shift” means the performance of an artificial intelligence model during deployment in the event real-world data differs statistically from the historical data on which such artificial intelligence model was trained.
Breach of security. The term “breach of security” means the unauthorized access, acquisition, disclosure, or use of any data or system that compromises the security, confidentiality, or integrity of any information or technology maintained by an agency.
Calibration. The term “calibration” means ensuring predictive confidence matches actual probabilities.
Data provenance. The term “data provenance” means the origins of the data on which an artificial intelligence model is trained and whether such artificial intelligence model was tested based on holdout data.
Determinism. The term “determinism” means the ability of an artificial intelligence model to always produce the exact same output when given the exact same input.
Holdout data. The term “holdout data” means a portion of data deliberately kept separate from an artificial intelligence model during its training and tuning.
Latency. The term “latency” means the total time delay between a user submitting an input to an artificial intelligence model and such artificial intelligence model generating a complete response.
Shut-down capability. The term “shut-down capability” means the technical capability for a human operator to cause an artificial intelligence model to temporarily or permanently stop functioning.
Third-party validator. The term “third-party validator” means a person, other than an affiliate of a developer of an artificial intelligence model, who is retained, contracted, or otherwise engaged by a developer of an artificial intelligence model to validate such artificial intelligence model in accordance with section 20-884 and based on the topics set forth in subdivision a of section 20-885.
Throughput. The term “throughput” means the amount of material, data, or work that passes through an artificial intelligence model in a specified amount of time.
§ 20-884 Third-party validation and shut-down capability. It shall be unlawful to market, offer for sale, sell, or deploy an artificial intelligence model unless:
1. Such artificial intelligence model has been validated by a third-party validator based on the topics set forth in subdivision a of section 20-885; and
2. Such artificial intelligence model includes a shut-down capability.
§ 20-885 Third-party validation; assessments, certification, disclosure. a. Assessments. A third-party validator shall conduct validation of an artificial intelligence model by performing assessments of the following topics:
1. Task performance, based on the criteria of accuracy, precision and recall, calibration, and behavior under distribution shift;
2. Determinism;
3. Latency and throughput;
4. Data provenance;
5. Disparate impact on, or bias against, individuals in protected classes as identified in chapter 1 of title 8 of this code;
6. Whether such artificial intelligence model handles data lawfully and securely and adheres to informed consent principles;
7. Safety, based on the criteria of the existence and functionality of a shut-down capability and the level of risk that deployment of such artificial intelligence model poses: (i) to human safety; (ii) of property damage or loss; and (iii) of a breach of security; and
8. Any other topics and criteria deemed relevant by the director of the office of cyber command.
b. Certification and disclosure of interests. Based on a validation a third-party validator conducts under subdivision a of this section, such third-party validator shall provide a certification and disclosure to the developer of an artificial intelligence model that has retained, contracted, or otherwise engaged such third-party validator, and to the office of cyber command, that set forth:
1. Whether such artificial intelligence model has been validated and accordingly is appropriately positioned for deployment; and
2. Any financial or other interest such third-party validator may have in such artificial intelligence model or its developer.
§ 20-886 Rulemaking. The director of the office of cyber command shall promulgate rules regarding:
1. Any additional topics or criteria deemed relevant by the office of cyber command for assessment under subdivision a of section 20-885;
2. Requirements concerning the form, content, and manner of submission to the office of cyber command of the certification and disclosure required under subdivision b of section 20-885; and
3. Qualifications for third-party validators.
§ 20-887 Penalties. a. A person that violates section 20-884 is liable for a civil penalty of $25,000, per instance of marketing, offering for sale, sale, or deployment of an unlawful artificial intelligence model.
b. A person that falsifies a validation conducted under subdivision a of section 20-885 is liable for a civil penalty of $25,000, on a per instance basis.
c. A person that violates a provision of this subchapter other than section 20-884 is liable for a civil penalty of up to $25,000.
§ 20-888 Enforcement. a. Civil penalties authorized under section 20-887 may be recovered in a proceeding before an administrative tribunal within the jurisdiction of the office of administrative trials and hearings. Administrative summonses returnable to such tribunal for violations of this subchapter may be issued by the office of cyber command or by an agency designated by the director of the office of cyber command.
b. The corporation counsel may bring a civil action on behalf of the city in any court of competent jurisdiction to recover any or all of the following:
1. Civil penalties authorized under section 20-887;
2. Injunctive relief to compel action under this subchapter or to restrain or enjoin any action in violation of this subchapter;
3. Attorneys’ fees and costs, and such other remedies as a court may deem appropriate.
§ 2. This local law takes effect 180 days after it becomes law.
NAW
LS #26835
9/25/2026 11:55 AM