New York State Court System Releases AI Policy

Last week, the New York State Unified Court System (UCS) unveiled its new policy on the use of generative artificial intelligence (AI) to perform various tasks.  

The policy, the result of a year-and-a-half of development, was initiated by Chief Administrative Judge Joseph A. Zayas, who formed a committee on the topic in April 2024.

The new policy has now been released. The seven-page document outlines the UCS’s stance on AI and establishes guardrails for its use. According to the UCS, this is a first-of-its-kind policy, and it hopes to enable the use of AI while minimizing the risks that come with its use.

To be clear, this policy only applies to UCS employees. It does not impact those filing with the courts. Still, it means that this binds judges and nonjudicial employees of the UCS, affecting decisions, policy memos and more.

However, as with most things, the devil is in the details. While the policy overall has decent guiding principles, it makes some bizarre decisions that may create new problems if they are not careful.

What the Policy Says

The first half of the document is primarily an overview of what generative AI is and the potential hazards of using it. The report acknowledges explicitly that AI fabricates information, may harbor biases, and can pose a risk when handling confidential information.

To that end, the policy has three tentpoles for the appropriate use of AI.

1. AI should never serve as a substitute for human judgment.
2. Private information should never be fed into a public AI system.
3. Employees should avoid using AI if it violates their ethical responsibilities. This includes issues of bias or prejudice.

To achieve that, the policy requires employees to undergo AI training before using any AI systems. The policy also forbids the use of paid AI systems, other than those provided by the UCS.

The policy also restricts which AI systems employees can use. The report outlines six AI systems employees can use, five of which are hosted either locally or in a space under UCS control. The only publicly available system permitted is ChatGPT, though paid subscriptions are prohibited.

Of the AI systems permitted, four are owned by Microsoft and include Microsoft Azure AI Services, Microsoft 365 CoPilot Chat, Microsoft 365 CoPilot and GitHub CoPilot for Business and Enterprise.

The remaining system is Trados Studio, which is used explicitly for translations.

However, what is most interesting about this policy is what it’s missing, not what’s in it.

Unusual Choices

On the surface, much of the policy makes sense. Clearly, the UCS has a contract with Microsoft to provide services. It makes sense to limit AI usage to those systems. Not only are they private models, but they are under the direct control of their Division of Technology and Court Research (DoTCR).

If something goes awry, they quickly disable or reduce access to these systems. They can also more easily spot abuse and misuse of those AI systems.

So why open employees up to ChatGPT at all? Not only is the model public, making it a potential risk to confidential information, but it’s also constantly evolving. The UCS won’t be able to control how employees use it, if at all.

Given the hazards of AI, which the policy acknowledges, this is opening the door to those exact issues. With the other AI systems, the UCS has a degree of control over usage, output, and data submitted to it; there are no such limits when an employee uses ChatGPT on a free account.

However, the report is missing one critical component: any discussion of enforcement. Will the DoTCR block access to unpermitted AI systems? How will they spot any unauthorized AI usage? What will the repercussions be?

A policy without enforcement is all but meaningless. With no indication of how this will be enforced, it remains to be seen if it will have any impact at all.

Bottom Line

As a skeptic of AI, I would have preferred their policy to be “don’t use AI.” In the legal field, the stakes are just too high (especially for judges), and AI’s tendency toward bias and fabrication makes it an untrustworthy tool in this space.

However, such a stance is likely impractical. AI isn’t just a tool that one has to seek out and use. It’s being baked into the products workers use every day. Microsoft is an excellent example of this.

With that in mind, this policy gets a great deal right. It makes it clear that the human is ultimately responsible for the completed work, regardless of how it was created. It mostly limits AI usage to private systems to ensure confidentiality. Finally, it requires training before an employee can use AI in their work.

While I’m still befuddled by the inclusion of free ChatGPT and wish there had been more information about enforcement, the policy seems grounded in today’s reality. It aims to mitigate the worst issues associated with AI use.

Misgivings aside, this is far from the worst AI policy that I’ve reviewed. This policy at least acknowledges the hazards of AI and is more about giving guidance to those who want to use it, rather than trying to force it on employees.

It’s a low bar, but it’s one that many companies and organizations are failing to clear.