As I have reported here, the legal technology company Clio is making a big push into the judiciary. It wants to develop AI tools to make judges\u2019 jobs easier, and it has even trademarked the phrase \u201cjudiciary-grade AI.\u201d To that end, Clio has made two key hires: Pablo Arredondo, the Casetext cofounder who joined Clio [\u2026]<\/p>\n
As I have reported here, the legal technology company Clio is making a big push into the judiciary. It wants to develop AI tools to make judges\u2019 jobs easier, and it has even trademarked the phrase \u201cjudiciary-grade AI.\u201d<\/p>\n
To that end, Clio has made two key hires: Pablo Arredondo, the Casetext cofounder who joined Clio<\/a> as senior vice president, judiciary, to lead the company\u2019s development of technology for the courts, and Bridget M. McCormack, the former Michigan Supreme Court chief justice and American Arbitration Association president, who is now general manager, judiciary, where she will lead market and commercial strategy for the judiciary business.<\/p>\n Clio also acquired <\/a>Learned Hand, a startup that builds AI exclusively for judges and courts, giving it an existing product already in use in a number of courthouses to build on.<\/p>\n But even as it does all this, a key question is whether courthouse doors are even open to Clio or any other AI provider? For that matter, are judges even permitted to use AI?<\/p>\n Turns out the answer varies widely by state, with at least one state outright banning AI use by its judges, and others having either more open policies or no policies at all.<\/p>\n The state that bans AI use by judges happens to be the one I live in \u2013 Massachusetts. Its policy, the most restrictive in the country, bans the use of AI for legal research, legal writing or other legal work.<\/p>\n The ban applies to judges, clerks, law clerks, interns, and other court employees and contractors.<\/p>\n The Massachusetts Supreme Judicial Court adopted the guidelines, \u201cInterim Guidelines for use of Generative AI,\u201d in November 2025. The guidelines say that they \u201cwill be refined and modified as more becomes known about Generative AI.\u201d A check last week with the court confirmed that there have been no modifications.<\/p>\n Curious about the policies in other states, I enlisted ChatGPT to help me conduct a 50-state survey, and then had Claude double-check the results. I also performed my own research. The results are below.<\/p>\n If you know of additions, changes or errors in anything below, please let me know (ambrogi-at-gmail.com).<\/p>\n The full findings are in the chart below. But first, a few comments and notes.<\/p>\n First off, to be clear, this survey does not cover policies directed solely at attorneys or litigants. The focus is solely on policies directed toward judges, magistrates, law clerks, court administrators, and the like.<\/p>\n It encompasses four categories of policies:<\/p>\n An important distinction is between using AI to assist judicial work and using AI to exercise judicial judgment. Some states explicitly authorize AI-assisted research and drafting while prohibiting judges from delegating decisions to AI. Others restrict the tools or circumstances in which judicial employees may use AI at all.<\/p>\n One clear conclusion from the survey is that there is no single national approach to judicial AI use. Rather, the policies fall along a spectrum, ranging from detailed rules to broader ethical statements to nothing at all.<\/p>\n As noted, Massachusetts is the most restrictive, prohibiting gen AI for legal research, legal writing, and other legal work. It permits the use of AI for administrative tasks.<\/p>\n By contrast, New York takes a more permissive approach. Its policy allows approved AI tools under specified conditions, and its September 2026 ethics opinion expressly recognizes using private AI to summarize judicial proceeding transcripts.<\/p>\n Rhode Island is an example of an approach that centers on ethics. Its detailed judicial guidance explores specific problems presented by AI, but the Supreme Court expressly makes the guidelines advisory rather than binding.<\/p>\n But there is one point on which these policies agree, which is that judges cannot delegate their judgment to AI. While AI may assist a judge, it may not serve as the decisionmaker.<\/p>\n Even when policies allow AI-generated research, summaries, or drafting, judges retain responsibility for independently understanding the record, applying the law, and deciding the case.<\/p>\n Another issue that comes up frequently is the need for verification of legal citations and factual assertions. These obligations are especially important given the risk of hallucinations, or even of hard-to-spot inaccuracies.<\/p>\n Some policies distinguish between public AI systems and private, approved or specially secured systems. Policies commonly prohibit judges and employees from entering nonpublic case information, internal judicial deliberations, sealed material, or personally identifying information into a public chatbot.<\/p>\n Some states \u2013 including Arizona, New York, Pennsylvania, South Carolina, Utah and Virginia \u2014 have adopted controls over which tools court personnel may use. Several also require training, supervisory permission, or disclosure of an employee\u2019s AI use to the judge for whom the employee works.<\/p>\n In the end, the biggest question is not whether judges may use AI for scheduling, meeting notes or routine correspondence. Rather, it is whether they may use AI to analyze evidence, summarize testimony, conduct legal research, draft opinions, or assist with factual findings.<\/p>\n With Clio moving into the judiciary market, others are sure to follow. As they do, distinctions about how judges and court personnel may or may not use AI become increasingly important.<\/p>\n This table reviews publicly available state court policies, rules, administrative directives, and guidance governing the use of generative AI by judges and judicial personnel as of Oct. 30, 2026. As I said, I used ChatGPT to perform the research and compile the table, then had Claude check the results, and then supplemented it with information I found on my own.<\/p>\n Note that \u201cnot identified\u201d does not necessarily mean no policy exists, just that none was found. If you see gaps or mistakes, please let me know and I will update the chart.<\/p>\n As a reminder:<\/p>\nSome Notes on the Findings<\/h3>\n
\n
No National Consensus<\/h3>\n
State-by-State Findings<\/h3>\n
\n