New York Understanding Artificial Intelligence Act proposal (S10733)
AI-generated text. This page was generated using artificial intelligence.
Understanding Artificial Intelligence Act (proposed)[1]
New York, United States[1]
Proposed state legislation[1]
S10733 (2025–2026); proposed General Business Law Article 45-C, §§ 1550–1552[1]
Introduced November 6, 2026 by Senator Rachel May; referred to Senate Rules[2]
Would take effect immediately upon enactment; pending proposal[1]
November 8, 2026 (complete introduced text and Senate status)
Overview
S10733 is a New York proposal for developer liability for certain injuries caused by covered AI models to people who are not users. Senator Rachel May introduced it on November 6, 2026; the Senate record lists referral to Rules. The bill would add the Understanding Artificial Intelligence Act to the General Business Law.[2][1]
Scope
The covered-model definition combines a specified training-computation threshold with a training-compute cost exceeding $100 million, calculated using average cloud-compute market prices at the start of training as reasonably assessed by the developer. A developer is the person performing the initial training that meets the specified computation and cost requirements. Fine-tuning is defined as adjusting trained model weights through exposure to additional data.[1]
Proposed liability and causation
Except for defamation claims, developers would be strictly liable regardless of their degree of care for injuries to non-users that satisfy the actual-harm element of ordinary negligence, subject to both of these conditions:[1]
- The injuries must be factually and proximately caused by model conduct that would satisfy negligence, an intentional tort or a crime if undertaken by an adult human of sound mind.
- Neither the model user nor an intermediary that fine-tuned, scaffolded or otherwise modified the model must have intended or reasonably anticipated that conduct.[1]
For torts requiring a particular mental state, the bill would create a rebuttable presumption if the fact-finder concludes, by a preponderance of the evidence, that a natural person taking similar actions under similar circumstances would have had that state. It specifies the instruction when evidence tending to rebut the presumption is presented and would disallow a defense based solely on AI systems being incapable of mental states.[1]
Affirmative defenses
A developer could establish an affirmative defense by showing that the model met the standard of care applicable to humans performing the same function when the injury-causing conduct occurred. A separate capabilities-failure defense would apply where the model fell short of the user's intended or reasonably anticipated performance, but its conduct would not satisfy negligence, an intentional tort or a crime if performed by an adult human of sound mind.[1]
Status and effective date
The Senate record identifies the measure as in Rules following its November 6 referral. Section 2 proposes immediate effect upon enactment.[2][1]
Related articles and coverage
References
- ↑ 1.00 1.01 1.02 1.03 1.04 1.05 1.06 1.07 1.08 1.09 1.10 1.11 New York Senate, S10733 introduced print, November 6, 2026, complete two-page bill, sections 1–2.
- ↑ 2.0 2.1 2.2 New York State Senate, S10733, 2025–2026 session, sponsor, current status and November 6, 2026 action; checked November 8, 2026.