Artificial Intelligence and Civil Justice: U.S. Practice, Policy, and Principles^†

Introduction

In the coming years, artificial intelligence (AI) will change the legal profession and the civil justice system significantly and irrevocably. While AI may often be useful and even beneficial, certain guardrails must be established to ensure the quality of justice is not diminished.

When determining the appropriate boundaries of national law and practice, it can be helpful to consider how other jurisdictions approach the issue at hand. The International Academy of Comparative Law (IACL) has undertaken precisely such a study regarding AI, with findings to be published in an edited volume.¹ The current analysis constitutes an excerpt from the national report for the United States.

This Report follows the basic structure established by the general rapporteurs for the IACL comparative study, although the responses have been edited for space. The discussion begins in Section I with a general overview of AI in the United States, including both definitions and strategic initiatives. Next, Section II covers the legal and regulatory framework for AI in U.S. civil justice systems, addressing governance concerns both generally and specifically with respect to the judiciary. Section III focuses on key issues in civil justice, including an overview of existing case law; a discussion of AI’s role, benefits, and risks; the extent to which the United States allows fully automated decision-making; and transparency and accountability of AI systems. In Section IV, the Report addresses current or planned models of AI in civil justice and examines AI systems involving various types of court activity. This section also discusses capacity-building for judges and lawyers in the area of AI as well as public perception and trust of AI in civil justice systems. After reviewing economic and access to justice concerns in Section V, the Report provides an overview of future development and challenges in Section VI.

I. General Overview of AI in the United States

A. Definitions

Definitions are important in the world of AI. Processes vary widely and include such diverse technologies as rule-based AI, generative AI, and retrieval-augmented generation (RAG).² Each of these mechanisms gives rise to different practical and policy concerns, though most laws, policies, and guidelines do not distinguish among the different systems.

While there is no single definition of AI that is used consistently throughout the United States,³ one influential definition indicates that AI is a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations or decisions influencing real or virtual environments. Artificial intelligence systems use machine and human-based inputs to-

• (A) perceive real and virtual environments;

• (B) abstract such perceptions into models through analysis in an automated manner; and

• (C) use model inference to formulate options for information or action.⁴

State-level definitions vary significantly,⁵ although it is unclear whether and to what extent those variations will survive in the wake of Executive Order 14,365, which seeks to limit “excessive State regulation” of AI.⁶ At the time of writing, it appeared that the Executive Order was likely to be challenged on constitutional grounds, though no such action had yet been brought.⁷

B. Strategic Initiatives on AI and Digitalization

1. Federal Initiatives

a. Judicial Initiatives

The federal judiciary has engaged with AI at both the national and local levels. The most critical work at the national level is being done by the Judicial Conference.⁸ The Judicial Conference appears to have first considered AI in early 2024, when the Committee on Court Administration and Court Assistance reported on the emerging use of AI in preparing judicial submissions and increasing efficiency in court administration and the Committee on Federal-State Jurisdiction discussed a report from chief justices of various state courts concerning the effect of AI on the courts and the practice of law.⁹ The next year, the Judicial Conference issued a five-year strategic report that reflected an intent to “[e]stablish an AI governance framework to guide responsible adoption of AI and to manage risks and challenges presented by these new and evolving technologies.”¹⁰

The Judicial Conference’s Advisory Committee on Evidence Rules (Evidence Committee) has been particularly active with respect to AI.¹¹ For example, the Evidence Committee has considered whether to adopt a new Rule 707 on machine learning and a new Rule 901 on digital replicas, also known as deepfakes.¹² While the new rule on deepfakes remained in committee at the time of writing, a public consultation regarding Rule 707 was released in August 2025, with the public comment period concluding in February 2026.¹³ Other parts of the federal government, most notably the U.S. Copyright Office, have also taken steps to address deepfakes.¹⁴

The federal judiciary has also addressed AI through rules and orders promulgated by individual federal courts and judges.¹⁵ Several courts have cautioned litigants about the use of AI and noted the need to confirm and verify the accuracy of all pleadings and submissions.¹⁶ Individual judges have also adopted standing orders relating to the use of AI by litigants, although no known order describes whether and to what extent the judge can or will use AI.¹⁷ Instead, orders typically require litigants and/or counsel to ensure submissions are accurate by personally reviewing any AI-generated content.¹⁸ Some orders include explanations about why AI in judicial settings is problematic, perhaps for the benefit of pro se litigants who might not otherwise be aware of the challenges of AI in legal settings.¹⁹

b. Executive branch initiatives

The executive branch has addressed AI through a series of executive orders beginning in 2019.²⁰ Several orders note that the use of AI must not only comply with the Constitution but must also maintain values that are central to civil justice, including public trust and the protection of privacy, civil rights and civil liberties.²¹ However, recent executive orders suggesting that AI is not being implemented in a neutral manner²² may themselves reflect discriminatory intent or practices.²³

One executive branch initiative, entitled “Winning the Race: America’s AI Action Plan,” ostensibly gives courts and law enforcement agencies certain tools needed to address challenges associated with AI, particularly those relating to the use and authentication of evidence.²⁴ For example, the plan calls for the development of formal guidelines regarding deepfakes in various settings.²⁵

Federal adjudicatory bodies embedded in administrative agencies or executive branch departments are also considering the effect of AI on their work.²⁶ Thus, the Office of Management and Budget (OMB) has required any entity using AI in a rights-impacting function to assess and mitigate algorithmic discrimination; provide notice to individuals of adverse AI-based decisions and provide opportunities for review by or appeal to human decision-makers; and offer opt-outs or human alternatives where practicable.²⁷

The work of the OMB has affected other executive agencies as well. For example, the Department of Justice (DOJ) has created a plan for implementing a comprehensive AI governance program that is consistent with OMB guidelines.²⁸ The DOJ also publishes an AI Use Case Inventory describing the types of AI tools–including those used in litigation–that are used by the Department.²⁹

2. State initiatives

Space restrictions make it impossible to provide a fifty-state survey of state initiatives concerning AI, but states are actively working to address the challenges in this field. For example, a number of states address AI through local rules of court³⁰ or rules adopted by state supreme courts regulating the bar.³¹ Individual state court judges have also adopted rules on counsel’s use of AI.³² California has perhaps gone the farthest by requiring any state court that does not prohibit the use of generative AI to adopt a policy outlining the permitted uses of generative AI by (1) court staff (for any purpose) and (2) judicial officers (for non-adjudicative tasks).³³ Adjudicative tasks were excluded to respect judicial independence.³⁴

II. Legal and Regulatory Frameworks for AI in Civil Justice

A. General Legal Framework

The United States faces six major barriers when it comes to regulating AI: (1) the speed with which technology is outpacing legislation; (2) a lack of AI expertise among policymakers; (3) regulatory capture; (4) gridlock at the political level; (5) antiquated governance structures, and (6) inherent difficulties in understanding the complexity of AI.³⁵ These factors, combined with the traditional U.S. preference to regulate ex post through litigation rather than ex ante through legislation, make it unlikely that the United States will develop comprehensive AI regulation in the short term, particularly at the federal level.³⁶ Indeed, of the hundreds of bills on AI that have been introduced in Congress since 2017, only 30 have been enacted.³⁷ State legislatures have been more successful in enacting regulations concerning AI,³⁸ although it is unclear how many of these provisions will continue in force should Executive Order 14,365 pass constitutional muster.³⁹

At this point, the most influential regulatory instrument in the United States is Formal Opinion 512, which was promulgated by the American Bar Association (ABA) Standing Committee on Ethics and Professional Responsibility.⁴⁰ Formal Opinion 512 is only binding in those jurisdictions where it has been formally adopted but has provided a template for many state and local organizations’ approach to AI.⁴¹

Formal Opinion 512 requires attorneys who use generative AI to do so in a way that is consistent with existing ethical obligations.⁴² For example, the requirement that lawyers provide competent representation to clients means that a lawyer must not only have a reasonable understanding of the capabilities and limitations of any AI tool that the lawyer might use but must also independently review and verify the accuracy of the tool’s output.⁴³ Ethical obligations regarding client confidentiality mean that a lawyer must consider the risk that the use of AI will lead to the disclosure of confidential information.⁴⁴

Formal Opinion 512 does not prohibit the use of AI but does requires lawyers to disclose any use of AI to both courts and clients.⁴⁵ Lawyers with managerial and supervisory responsibilities must ensure that all subordinates comply with ethical guidelines regarding AI, regardless of whether the subordinates are lawyers.⁴⁶ Lawyers must also explain to clients how deploying AI will affect fees before using those tools.⁴⁷

While judicial behavior is typically regulated by judicial codes of conduct, many of the principles reflected in Formal Opinion 512 have been extended to cover judges.⁴⁸ For example, Illinois expressly confirms that existing ethical rules for both lawyers and judges apply to cases involving AI.⁴⁹ However, Illinois diverges from Formal Opinion 512 in that Illinois does not require disclosure of the use of AI in judicial pleadings.⁵⁰

California also echoes Formal Opinion 512 in provisions relating to judicial use of AI.⁵¹ According to the California Standards of Judicial Administration:

A judicial officer using generative AI for any task within their adjudicative role:

1. Should not enter confidential, personal identifying, or other nonpublic information into a public generative AI system … .

2. Should not use generative AI to unlawfully discriminate against or disparately impact individuals or communities based on [­various] classification[s] protected by federal or state law.

3. Should take reasonable steps to verify that generative AI material, including any material prepared on their behalf by others, is accurate, and should take reasonable steps to correct any erroneous or hallucinated output in any material used.

4. Should take reasonable steps to remove any biased, offensive, or harmful content in any generative AI material used, including any material prepared on their behalf by others.

5. Should consider whether to disclose the use of generative AI if it is used to create content provided to the public.⁵²

B. AI Governance in the Judiciary

1. Soft-Law Instruments

Judicial education organizations have issued several soft-law instruments relating to AI in the judiciary. For example, the Federal Judicial Center’s (FJC) 2023 publication, An Introduction to Artificial Intelligence for Federal Judges, discusses how judges might consider AI evidence under the Federal Rules of Evidence and their state-law equivalents.⁵³ The guide also considers how AI-generated evidence might vary depending on the nature of the dispute and how AI-generated evidence might affect various constitutional protections.⁵⁴

The National Center for State Courts (NCSC) has published various “bench cards” dealing with AI.⁵⁵ These short-form judicial guides outline issues judges must consider when presented with both acknowledged and unacknowledged AI-generated evidence.⁵⁶

The Sedona Conference has also published guidelines regarding the use of AI by judges and judicial staff.⁵⁷ The guide provides an AI use case inventory (i.e., a list of the types of judicial tasks that AI is competent to handle) and outlines limitations and best practices associated with the use of AI by judges and court staff.⁵⁸

State-level groups have also promulgated soft-law instruments on AI.⁵⁹ Interestingly, the Delaware state policy on AI has expressly forbidden the delegation of decision-making functions to AI tools.⁶⁰

Soft law also exists regarding the use of AI in mediation, which can arise in the United States on a voluntary, contractual or court-ordered basis. Some instruments have a purely domestic focus,⁶¹ while others provide an international perspective.⁶²

Arbitration is another important form of civil justice in the United States and may be an early adopter of AI due to the consensual nature of the process.⁶³ The most detailed soft-law instrument in the arbitral realm comes from the Silicon Valley Arbitration & Mediation Center (SVAMC). The SVAMC guidelines were promulgated for the use of arbitrators, counsel, and parties and therefore address many of the same types of concerns as other hard- and soft-law instruments.⁶⁴ However, the SVAMC guidelines also explicitly discuss adjudicative concerns (such as the non-delegation of decision-making responsibility and respect for procedural fairness) that will be of interest to judges as AI begins to make more significant inroads into judicial processes.⁶⁵

2. Data Protection and AI

At this point, there is no comprehensive approach to data regulation and AI in the United States.⁶⁶ Data protection in federal litigation is addressed to some extent by Federal Rule of Civil Procedure 5.2, which allows redaction of sensitive information from court filings.⁶⁷ However, this rule is not well aligned to the kind of data protection issues involving AI. Formal Opinion 512 also discusses confidentiality but focuses more on traditional concerns about client confidentiality than on data protection per se.⁶⁸

Some soft-law guidance on this subject comes from the NCSC. While the NCSC does not provide any specific policies or implementation recommendations regarding data protection, it does note the importance of complying with any relevant laws.⁶⁹ The Sedona Conference has published similar admonitions about data protection issues relating to the use of AI in civil litigation.⁷⁰

The absence of dedicated legislation regarding data protection and AI means that prosecuting entities will have to be creative in how they address data breaches in civil litigation. One possibility involves the Federal Trade Commission Act (FTCA), which prohibits unfair or deceptive acts affecting commerce.⁷¹ The Federal Trade Commission (FTC) has begun to use the FTCA to prohibit unfair or deceptive practices relating to the use of AI.⁷² One such action involved DoNotPay, which claimed to act as “‘the world’s first robot lawyer’ and an ‘AI lawyer.’”⁷³ The matter was ultimately settled through a consent order.⁷⁴

Another route to recovery involves state laws concerning AI-assisted biometric systems. For example, In Re Clearview AI, Inc. Consumer Privacy Litigation involved the automatic collection, storage and use of biometric data violating several states’ privacy laws.⁷⁵ The ensuing class action was settled for $51.75 million.⁷⁶  State of Texas v. Meta Platforms, Inc. involved allegations that Meta violated several Texas laws by using Texans’ biometric data without their knowledge or permission.⁷⁷ Meta agreed to pay the State of Texas $1.4 billion to settle the matter.⁷⁸ Similar allegations in a separate case against Google led to a $1.375 billion settlement.⁷⁹

Data breaches can also arise through the use of AI in litigation itself. In Re: OpenAI, Inc. Copyright Infringement Litigation involved a consolidated class action relating to claims by various authors and copyright holders that OpenAI trained its large language model (LLM) on their copyrighted works and created infringing works in the LLM outputs.⁸⁰ During discovery, OpenAI allegedly deleted output log data, claiming a need to respect “not only user preferences and requests, but also ‘numerous privacy laws and regulations throughout the country and the world that also contemplate these type of deletion requests or that users have these types of abilities.’”⁸¹ The court subsequently ordered OpenAI to preserve any output log data that would otherwise be deleted, seemingly elevating the needs of the litigants over existing privacy laws and data protection regulations.⁸²

3. Security and Integrity of AI Systems

Although there do not appear to be any federal laws relating to the security and integrity of AI systems, some states have adopted relevant provisions. For example, Colorado requires AI developers to prevent “algorithmic discrimination,” whereby the use of an AI system would result in “an unlawful differential treatment or impact that disfavors an individual or group of individuals” who are protected as a matter of state or federal law.⁸³ Texas prohibits the development or deployment of certain types of AI, but only in cases where there is an intent to unlawfully discriminate against individuals protected under state or federal law.⁸⁴ Notably, these and similar statutes are in danger of repeal pursuant to Executive Order 14,365, which seeks to deregulate the AI industry at the state level.⁸⁵

Most concerns relating to the security and integrity of AI systems in civil justice involve the impact of AI on the integrity of evidence introduced at trial. For example, Washington v. Puloka considered whether to admit digitally enhanced video into evidence following claims that AI had created false details in a particular image.⁸⁶ Following expert testimony, the court refused to admit any AI-enhanced videos into evidence on the grounds that the technology in question was neither peer-reviewed nor reproducible by the forensic video analysis community; no known state or federal court had approved the use of AI-enhanced video in any criminal or civil proceeding; and the use of the video would lead to confusion of the issues and unfair prejudice.⁸⁷ Though Puloka was a criminal case, the court’s approach would appear relevant to civil disputes as well.⁸⁸

Ferlito v. Harbor Freight Tools USA, Inc., also involved AI in the litigation process.⁸⁹ In this case, an expert witness confirmed his conclusions about how best to affix an axe head to a handle by running a query through ChatGPT.⁹⁰ The court allowed the expert opinion to stand, holding that there was “little risk” that use of AI impaired the expert’s methodology because AI was introduced only after the expert had formed his conclusions.⁹¹

As these examples show, concerns about AI-enhanced evidence are real and rising. As a result, the Judicial Conference’s Evidence Committee is considering amending the Federal Rules of Evidence to include a new Rule 901(c) on Potentially Fabricated Evidence Created by Artificial Intelligence, colloquially known as deepfakes.⁹² The 2025 version of the proposed rule states:

1. Showing Required Before an Inquiry into Fabrication. A party challenging the authenticity of an item of evidence on the ground that it has been fabricated, in whole or in part, by generative artificial intelligence must present evidence sufficient to support a finding of such fabrication to warrant an inquiry by the court.

2. Showing Required by the Proponent. If the opponent meets the requirement of (1), the item of evidence will be admissible only if the proponent demonstrates to the court that it is more likely than not authentic.

3. Applicability. This rule applies to items offered under either Rule 901 or 902.⁹³

Under the 2025 version of the proposed rule, a party opposing entry of material into evidence on the ground that it has been fabricated must provide sufficient information to allow “a reasonable person to find that the item has been fabricated in whole or part by the use of generative AI.”⁹⁴ The burden then shifts to the proponent of the evidence to prove the materials are authentic, more likely than not.⁹⁵ As useful as this approach seems in theory, scholars have identified access to justice concerns relating to the inability of some litigants to afford experts who can recognize and combat deepfakes.⁹⁶

The Evidence Committee is also considering a possible new Rule 707 on machine-generated evidence.⁹⁷ That rule reads:

When machine-generated evidence is offered without an expert witness and would be subject to Rule 702 if testified to by a witness, the court may admit the evidence only if it satisfies the requirements of Rule 702(a)-(d). This rule does not apply to the output of simple scientific instruments.⁹⁸

4. Liability for AI-Related Errors

Over the last few years, various types of substantive claims have been brought for AI-related errors and improprieties. These actions range from breach of copyright⁹⁹ and incitement to suicide, self-harm, or murder¹⁰⁰ to deceptive practices¹⁰¹ and employment discrimination.¹⁰² As important as these matters are, this subsection focuses on errors relating to judicial proceedings.

At this point, there does not appear to be any state or federal legislation addressing liability for AI-related errors in the civil justice system. Indeed, some legislators take the view that any such regulation should come from the judicial branch, not the legislative branch.¹⁰³ This position was enunciated by the Chair of the Senate Committee on the Judiciary following reports that two federal judges—Judge Henry T. Wingate of the U.S. District Court for the Southern District of Mississippi and Judge Julien Xavier Neals of the U.S. District Court for the District of New Jersey–had engaged in egregious judicial misconduct by relying inappropriately on AI.¹⁰⁴ Though the Judicial Conference has begun to consider issues relating to the use of AI, it has not yet established any formal guidelines, rules, or causes of action.¹⁰⁵

Despite this lacuna, some general legislation may allow recovery in some cases. For example, the California Civil Code indicates that “In an action against a defendant who developed, modified, or used artificial intelligence that is alleged to have caused a harm to the plaintiff, it shall not be a defense, and the defendant may not assert, that the artificial intelligence autonomously caused the harm to the plaintiff.”¹⁰⁶

III. AI in Practice: Specific Issues in Civil Justice

A. Case Law

U.S. state and federal courts are issuing an increasing number of rulings on the use of AI in civil litigation. Again, this subsection focuses on matters relating to procedural concerns rather than violations of the substantive law.¹⁰⁷

The first wave of AI-related jurisprudence involved hallucinated citations contained in legal filings. Some courts characterized such malfeasance as a violation of the duty under Rule 11 of the Federal Rules of Civil Procedure and state analogues to verify the accuracy of submissions, whereas other courts framed the behavior as “frivolous,” an abuse of process, or a failure of the attorney to provide the client with competent representation.¹⁰⁸

The first of these cases involves the now-landmark decision in Mata v. Avianca, Inc.¹⁰⁹ While recognizing that “there is nothing inherently improper about using a reliable artificial intelligence tool for assistance,” the court held that attorneys are subject to a “gatekeeping role … to ensure the accuracy of their filings.”¹¹⁰ The court outlined various harms that can arise from the misuse of AI, including (1) the waste of the opponent’s time and money as a result of the need to identify and expose AI-generated errors; (2) the distraction of the court from other important matters; (3) the client’s loss of arguments based on actual judicial precedents; (4) possible injury to the reputation of judges and courts named in the hallucinated material; (5) increased cynicism in society at large about the credibility of lawyers and the judicial system; and (6) the possibility that a future litigant might attempt to defy a judicial ruling by expressing doubt about the authenticity of that order.¹¹¹ The court imposed various sanctions on the offending attorney, including a $5,000 fine.¹¹²

In the years since Mata was handed down, the number of submissions involving hallucinations has continued to rise.¹¹³ As judges have become increasingly frustrated, they have increased the amount of fines, with the highest known amount set at $15,500 in December 2025.¹¹⁴

Questions have been asked about who is continuing to use AI improperly despite significant publicity about the consequences of doing so. According to one recent study, filings with hallucinations are typically submitted by plaintiffs rather than defendants and primarily come from solo practitioners or small firms (two to twenty-five lawyers).¹¹⁵ Furthermore, ChatGPT is the primary type of AI used.¹¹⁶

Because Rule 11 is directed to any party presenting a pleading or other paper to the court, attorneys are not the only ones who can be sanctioned for relying on unverified AI.¹¹⁷ Thus, in Kruse v. Karlen, an unrepresented party had his appeal dismissed following reliance on hallucinated cases,¹¹⁸ while Kohls v. Ellison saw an expert declaration excluded because the expert cited non-existent articles following the use of AI.¹¹⁹

The range of sanctions used to address problematic behavior involving AI is expanding beyond fines and rejected submissions.¹²⁰ Reliance on hallucinated cases and language can also constitute a breach of an attorney’s professional responsibilities, leading some courts to sanction lawyers by referring them to the relevant disciplinary body.¹²¹

Although some matters involving AI do not fall under Rule 11 and state analogues,¹²² courts can nevertheless address problematic behavior pursuant to the judge’s inherent power to control litigation.¹²³ Sanctions might include disqualifying an attorney from the matter at hand, referring an attorney to the relevant licensing authority, or requiring an attorney to provide a copy of the order containing the sanction to the attorney’s client and/or the judge in any other case where the attorney is appearing.¹²⁴

B. AI’s Role, Benefits and Risks

Conversations in the United States about the benefits of AI typically focus on AI’s ability to increase efficiency, thereby reducing the time and cost of litigation. Although providers of AI services often seek to quantify the amount of time or money that is saved by their products, this type of “corporate empiricism” may not comply with best practices in social science research.¹²⁵ Such data must therefore be regarded with skepticism until it can be confirmed by independent and objective analysis.

Efficiency arguments involving AI have proven problematic on other fronts as well. For example, courts and commentators have both noted that claims of time and cost savings may be exaggerated when viewed from a systemic perspective, since the burden of checking for hallucinations and other errors may simply be passed from the user on to other participants in the litigation process.¹²⁶ Efficiency-related claims also tend to ignore the environmental impact of AI, including the massive amounts of energy and water needed to train and maintain AI data centers.¹²⁷

Another commonly cited benefit of AI is its purported ability to increase access to justice. Again, this phenomenon is more nuanced than it may appear. Widespread use of AI, particularly by self-represented litigants, could open the floodgates of litigation, overwhelming the courts.¹²⁸ While one response would be to automate courts, either fully or partially, to speed response times, the “‘induced traffic’ phenomenon” suggests that efforts to increase judicial productivity will not reduce waiting times but will instead increase the number of cases filed.¹²⁹ Questions have also been raised as to whether unrepresented parties who rely on AI in their legal submissions could lose the rights and benefits associated with appearing pro se.¹³⁰

Perhaps the most frequently raised risk to AI involves technological concerns, such as hallucinations and algorithmic bias. However, AI also gives rise to various social concerns, such as the effect on AI on public confidence in the judiciary and protection of fundamental rights, as well as individual concerns, such as those relating to deskilling, cognitive bias, effect of AI on knowledge workers’ core tasks and motivation, etc.¹³¹

Discussions about AI’s effect on fundamental rights have focused primarily on privacy.¹³² However, algorithmic bias can violate the judicial duty to act fairly and impartially by creating outcomes that discriminate on the basis of racial, gender and other protected characteristics,¹³³ while judicial use of AI may infringe on the non-delegation doctrine and the due process right to a human decision-maker.¹³⁴

None of these issues are unsolvable, although mitigation efforts depend on the nature of the problem.¹³⁵ For example, the risk and effects of algorithmic bias can be reduced by cooperative dialogue between AI designers, policymakers, and members of the affected group.¹³⁶ Risks relating to the use of AI-generated evidence can be lessened through careful application of the relevant rules of evidence.¹³⁷

C. Fully Automated Decision-Making

Fully automated judicial decision-making is not yet a reality in the United States,¹³⁸ although AI already supports various aspects of the litigation process, such as intake, negotiation, and document review.¹³⁹ Remote (virtual) proceedings have become increasingly common since the COVID-19 pandemic,¹⁴⁰ and various forms of online dispute resolution (ODR) are available in a limited capacity in a small number of public sector settings.¹⁴¹

While no mandatory automated online courts exist in the United States, an increasing number of jurisdictions require parties to use online features at some point during the litigation process.¹⁴² For example, parties in Utah’s small claims court must file their matters through the court’s ODR system unless certain exceptions (such as hardship) apply.¹⁴³ The matter is first heard by a human “facilitator” who assists the parties in reaching settlement.¹⁴⁴ If settlement is not achieved, the matter goes to trial.¹⁴⁵

As courts consider what types of automation to adopt, they need to consider how those procedures might affect due process protections embodied in various state and federal constitutions and laws.¹⁴⁶ At this point, the primary protection involves ensuring that AI-generated information is reviewed by a human subject-matter expert.¹⁴⁷ Indeed, two federal judges who used AI to help write judicial pronouncements without reviewing the output sufficiently have been severely criticized.¹⁴⁸

Parties may choose to adopt fully automated decision-making through various forms of ODR. While ODR is primarily associated with private-sector disputes, it has been adopted in a few public-sector settings.¹⁴⁹ The most common public mechanism involves courts encouraging parties to engage in court-supported online mediation, although the mediators are humans rather than machines.¹⁵⁰ Appeals in ODR are permitted to the extent dictated by the companies and services that provide the platform in question.¹⁵¹ The ABA has published detailed guidelines on the appropriate use of ODR to encourage adherence to best practices.¹⁵²

D. Transparency and Accountability

Transparency in AI “reflects the extent to which information about an AI system and its outputs is available to individuals interacting with such a system–regardless of whether they are even aware that they are doing so.”¹⁵³ At this point, there are very few legal requirements governing algorithmic transparency, though the ABA adopted a resolution early in 2019 urging courts and lawyers to address matters involving bias and transparency.¹⁵⁴ In 2024, the National Institute for Standards and Technology issued non-binding advice regarding supervision and auditing of AI processes, consistent with the federal government’s then-existing focus on AI safety.¹⁵⁵ Since January 2025, the federal government has shifted its priorities regarding AI and now focuses more on deregulation and national security concerns than on safety.¹⁵⁶

One way to supervise or control the use of AI in judicial proceedings is through Rule 11 of the Federal Rules of Civil Procedure and its state analogues, since those rules require parties to verify the accuracy of all legal citations and filings.¹⁵⁷ Formal Opinion 512 could similarly be characterized as having an indirect regulatory effect, since Formal Opinion 512 requires lawyers to verify and disclose the use of AI in their submissions.¹⁵⁸

Other indirect measures of control may arise through substantive state laws. For example, Colorado regulates algorithmic discrimination that affects consumers.¹⁵⁹ Particular attention is paid to “high risk” AI systems, which are defined as systems that make or are involved in making a “consequential decision,” which is itself defined “a decision that has a material legal or similarly significant effect on the provision or denial to any consumer of, or the cost or terms of … [a] legal service.”¹⁶⁰ On its face, this law could affect the behavior of developers of legal AI by creating civil liability for wrongful actions.¹⁶¹

IV. Current or Planned Models of AI in Civil Justice

A. Existing AI Systems

AI has arguably been used by judges and lawyers for years, albeit without the “AI” nomenclature.¹⁶² For example, AI is incorporated at least to some extent in e-discovery platforms, legal research tools such as Westlaw Edge or Lexis + that suggest related legal authorities, and court scheduling systems that use AI to predict conflicts and optimize calendars.¹⁶³ Space restrictions preclude discussion of these early AI tools¹⁶⁴ as well as analysis of the ever-increasing number of AI products being marketed to practitioners.¹⁶⁵ Instead, the emphasis here is on the use of AI in and by courts.

Several jurisdictions use AI to assist with the initiation of lawsuits. For example, the County Clerk of Court’s office in Palm Beach, Florida, has been using an AI-driven system to automatically classify, index, and docket e-filed court documents since 2018.¹⁶⁶ Other Florida state courts have used AI to scan incoming filings for confidential data and redact sensitive information.¹⁶⁷

Courts also use AI to interact with the public through chatbot technology.¹⁶⁸ Court chatbots typically provide round-the-clock, sometimes multilingual, assistance to users, with human assistance available during office hours if the chatbot cannot resolve an inquiry.¹⁶⁹ Some bots also provide self-help resources for pro se litigants.¹⁷⁰

While most chatbots are only available online, the courts in Ottawa County, Michigan, offer a mobile robot kiosk that uses AI and voice recognition to assist visitors in the courthouse.¹⁷¹ Initially deployed as a “concierge,” the Court Operated Robotic Assistant (CORA) provides interactive maps and directions inside the courthouse, searchable court dockets and hearing schedules, and answers to frequently asked questions in both English and Spanish.¹⁷²

Courts are also using AI to help divert appropriate cases to alternative dispute resolution. For example, the Court of Common Pleas in Lancaster County, Pennsylvania, partnered with the NCSC and the American Arbitration Association (AAA) to create an AI-assisted screening system for the courts’ civil credit card debt diversion program.¹⁷³ The system handles repetitive administrative tasks that would otherwise be handled by court staff, thereby preserving judicial resources, though human supervisors oversee the AI outputs to ensure the cases are appropriately handled.¹⁷⁴

Courts also use AI during hearings, albeit to a limited extent. In 2017, Ohio State Court Judge Anthony Capizzi began using Watson AI to assist with his juvenile drug treatment docket.¹⁷⁵ Judge Capizzi–who has onlyfive to seven minutes with each juvenile defendant appearing on a typical drug treatment docket–uses AI to help sift through the 30-300 pages of paperwork associated with each individual case.¹⁷⁶ Judge Capizzi was still using the system in 2024,¹⁷⁷ and similar tools have been embraced by other specialized courts, including those involving veterans.¹⁷⁸

Courts are increasingly using AI in other aspects of the trial process. For example, some courts use AI-assisted speech-recognition programs to create transcripts of court proceedings.¹⁷⁹ Arizona has gone even farther and used AI avatars to announce case decisions and opinions.¹⁸⁰

As diverse as these various applications are, they all keep humans in the loop for quality control and ethical reasons.¹⁸¹ However, human intervention is in some cases discretionary, which could prove problematic given the unconscious cognitive distortion known as automation bias, which arises when individuals trust technology more than human intellect.¹⁸²

B. AI in Court Administration, Case Management and Court Registries

A number of guidelines have been published describing how AI can assist with court administration. (Interestingly, the National Association of Court Management took a somewhat Orwellian approach and had its guide drafted by AI.¹⁸³) The NCSC’s guide suggests that AI can be used to automate document classification, data extraction, and redaction to accelerate file processing time and reduce errors.¹⁸⁴ A number of courts have adopted measures consistent with this recommendation.¹⁸⁵

While predictive AI has not yet been used by courts to assist with case management, an AI-powered open data network called SCALES now allows users not only to extract data from the federal courts’ electronic access system (PACER) but also to transform and enrich that data to allow nuanced analysis.¹⁸⁶ While SCALES does not constitute predictive AI per se, it has increased transparency about judicial processes and outcomes in civil litigation.¹⁸⁷

Finally, courts in the United States have used AI to assist with ancillary tasks associated with court registers. For example, California’s Santa Clara County has used AI to remove racist elements from property records, consistent with a 2021 state order requiring counties to remove such language from their property records.¹⁸⁸ The tool scanned 5.2 million deed records to identify 7,500 deeds with racial covenants, saving 86,000 hours of work.¹⁸⁹

C. Capacity Building, Education, and Training

Law schools across the country are adopting new classes in AI as well as policies regarding the use of AI in graded assessments.¹⁹⁰ Commentators have also proposed reforming legal education on a more wholesale basis, based on the needs of the AI era.¹⁹¹ While law schools have attempted to adapt their curricula to respond to the needs of future practitioner, concerns exist that such reforms are often inconsistent with empirical research concerning the effect of AI on student learning and motivation.¹⁹²

Qualified lawyers are subject to continuing legal education (CLE) requirements, though details vary by state. While lawyers are typically free to choose how they will fulfil their CLE requirements, some states require attorneys to take a certain number of units involving cybersecurity and technology.¹⁹³ Individuals are free to exceed that number, and the massive surge of interest in AI from the legal community has led CLE providers to offer a wide variety of coursework on the subject.¹⁹⁴

Judicial education on AI is quite different. Federal judges and many state judges are not required to engage in any specialized education prior to or after their elevation to the bench.¹⁹⁵ Furthermore, there are no requirements that judges who voluntarily take classes focus on any particular subject.

While many judges choose to attend judicial education events offered by publicly funded organizations such as the FJC and the National Judicial College, other judges attend events offered by private organizations, including industry groups. Over the years, questions have been raised about the propriety of allowing judges to attend educational programs organized by members of the private sector, since it is unclear whether the information provided to the judges is entirely objective.¹⁹⁶ Similar questions may arise in cases involving industry-led forms of judicial education on AI-related matters.

Judicial education organizations also develop bench books and manuals on subjects of interest to judges. Many of these AI-related texts have been discussed elsewhere in this Report.¹⁹⁷

Most institutions offering judicial education also offer programming for judicial staff.¹⁹⁸ Some organizations have already developed role-specific training in AI.¹⁹⁹ However, training of court staff appears to be lagging behind training for judges and lawyers, as only 25% of court systems nationwide have offered their staff some type of training in AI.²⁰⁰

D. Public Perceptions and Trust

Some empirical research has been conducted on perceptions of AI in the justice system. According to one study, judges are primarily concerned about AI’s effect on normative concerns like fairness and legitimacy and are more receptive to AI when AI is presented as enhancing judicial skills and capabilities rather than merely as a means of increasing efficiency.²⁰¹ Judges have also been found to be more trusting of AI when it is used for legal research, drafting, and other types of administrative or support tasks.²⁰²

The most significant areas of concern for judges involve the use of AI in matters requiring judicial discretion and the effect of AI on judicial independence and de-skilling.²⁰³ On a more practical level, judges also worry about the possibility that AI could result in courts being overwhelmed with increased numbers of lawsuits.²⁰⁴

Lawyers, on the other hand, take a much more pragmatic view of AI, focusing primarily on AI’s ability to increase efficiency in drafting and research.²⁰⁵ However, practitioners have not adopted AI as quickly or as comprehensively as initially anticipated, suggesting a certain amount of unease about the new technology.²⁰⁶

Confidence in AI varies according to task.²⁰⁷ All individuals perceive AI as fairer during the information-gathering stage than in other stages, but lawyers consider AI in the decision-making stage to be less fair than laypeople do.²⁰⁸ Studies relating to the use of AI in mediation suggest that parties are more likely to accept mediator use of AI for preparatory tasks than for tasks used during the mediation or intended to benefit other parties or mediators in the future, as through the creation of an AI tool by the mediator that seeks to facilitate future mediations.²⁰⁹

V. Economic and Access-to-Justice Considerations

Reliable information about the economic benefits of AI on civil proceedings is difficult to compile, since the rise of “corporate empiricism” has led profit-motivated entities and industry actors to sponsor “studies” that yield desired outcomes, such as the conclusion that using the sponsoring organization’s goods or services will result in monetary savings.²¹⁰ Most of the information coming from reliable, non-interested sources is at this point speculative or anecdotal.²¹¹

AI has been touted as a means of increasing access to justice, in that unrepresented parties can use AI to help generate more persuasive arguments in court.²¹² However, concerns have been raised that AI (particularly high-quality legal AI) will not reach all individuals equally, with self-represented litigants, solo practitioners, and small firms being least likely to benefit from quality legal AI.²¹³ As a result, various proposals have been made about how to ensure equal access to justice in the AI era.²¹⁴ For example, proponents claim that AI itself can improve access to justice if free or reasonably priced AI programs are made widely available to facilitate self-representation²¹⁵ and if AI is properly calibrated to address the needs of different groups, including those who have been historically marginalized in the digital community.²¹⁶

Though the discussion is often framed in terms of access to justice, it is also useful to consider the issue from the perspective of digital exclusion.²¹⁷ When framed in this light, solutions include offering alternative (non-digital) communication methods; ensuring website accessibility, particularly for individuals with disabilities or with language barriers; partnering with community centers; hosting workshops on legal technology; creating easy-to-follow guides that assist with digital legal processes; and setting up telephone hotlines to provide technical assistance.²¹⁸ Other suggestions include offering legal services both online and offline and ensuring that any AI tool prompts individuals to speak with lawyers when necessary.²¹⁹

VI. Future Developments and Challenges

Like other countries, the United States is only beginning to grapple with the challenges of AI in its civil justice systems. While much remains in flux,²²⁰ several themes nevertheless seem to be emerging.

First, it is unclear whether AI will develop in the United States in a largely unregulated environment or in a diversely regulated one. While the United States is known for limiting formal regulation ex ante and allowing litigation–including litigation by “private attorneys general”—to act as a regulatory device ex post,²²¹ several states have nevertheless shown a willingness to regulate various aspects of AI to promote fairness and safety within the industry.²²² This type of regulatory diversity is often praised as reflecting a practical “laboratory” for legislatures,²²³ but Executive Order 14,365 suggests an intent to limit not only federal but also state regulation of AI.²²⁴ If upheld, Executive Order 14,365 will create an environment where individuals and institutions–including the courts–are vulnerable to problematic AI tools and processes.²²⁵ The legal profession has sought to safeguard certain aspects of civil justice through Formal Opinion 512 and state analogues, but those provisions do not address all possible ills arising from AI.²²⁶ Furthermore, the continuing number of cases with hallucinated citations shows that self-regulation–even of professionals–is not always effective.²²⁷

Second, AI will likely develop in a manner designed to promote personal rather than public goods. Although some forms of legal AI are clearly being created to promote the public interest,²²⁸ the majority of tools are being developed to maximize developers’ individual or institutional profit.²²⁹ In some cases, mechanisms designed for use by judges, lawyers, and others in the legal profession are developed by individuals with little or no legal training, giving rise to concerns about whether various procedural and substantive rights are receiving proper protection.²³⁰ The rise of corporate empiricism creates additional challenges, since industry-supported studies may dilute the impact of legitimate research into the advantages and disadvantages of AI in civil justice systems.²³¹

Third, U.S.-based academics and policymakers appear to be focused primarily on problems associated with the technical aspects of AI. Far less attention is being paid to concerns raised by social scientists about the effect of AI on the cognition and work patterns of knowledge workers or on the possibility that individuals can become addicted to generative AI to the same extent they can become addicted to substances and other digital technology.²³² It is also unclear whether systemic considerations, such as the effect of AI on public perceptions of the courts, are being adequately addressed by judges, legislators, or AI developers.²³³

Conclusion

The challenges of AI in civil justice are as real as the benefits. Every country is responding to those challenges in a different manner, based, in large part, on pre-existing legal and social norms.²³⁴ As such, it is unsurprising that the U.S. approach to AI, both generally and with respect to civil justice systems, is largely being driven by a capitalist ethos. Developers of AI are already seeking to maximize individual and institutional profit in a lucrative and largely deregulated market environment by aiming their first wave of innovations at practicing lawyers, who are seen as the biggest and wealthiest consumers of legal AI.²³⁵ It is expected that developers’ attention will soon turn to unrepresented parties, who make up for a lack of individual wealth with large numbers. Technology specifically aimed at judges and judicial staff will likely develop slowly, if at all, since courts have more limited budgets than law firms. Furthermore, judicial skepticism about the effects of AI on litigants’ rights and on judicial independence may slow acceptance of some forms of court-related AI. However, courts will be affected by general-use AI as well as legal AI that is used to facilitate court administration, legal research, and drafting.

Some observers might worry that the relative lack of AI-specific regulation will create an environment where injuries exist without a remedy. It is certainly true that plaintiffs and regulatory entities seeking legal recourse will have to rely on existing laws that may provide an imperfect fit for the injuries suffered. The paucity of specialized regulation may be most keenly felt in issues relating to transparency, since it may be difficult to establish a cause of action if it is unclear how a particular type of AI operates. Fortunately, several mechanisms exist that might mitigate this particular problem.

First, calls have already been issued noting the need for whistleblower actions involving the AI industry.²³⁶ While whistleblower actions may be the most obvious means of uncovering hidden wrongdoing, those cases require a principled and knowledgeable insider who is willing to step up and publicize an issue. Outsiders may not believe themselves capable of bringing such actions, since they lack the necessary knowledge.

Here, the second mechanism comes to the fore: discovery. Although many have bemoaned the broad scope of discovery under Rule 26 of the Federal Rules of Civil Procedure and state analogues, the robust nature of the U.S. discovery process may be used to good purpose by allowing litigants to overcome the secrecy surrounding various AI tools and procedures.²³⁷

Of course, discovery only arises pursuant to existing litigation, which brings us to the third means of mitigation: actions by “private attorneys general.”²³⁸ Such procedures might include class actions under Rule 23 of the Federal Rules of Civil Procedure and state analogues as well as civil claims brought under the Racketeer Influenced and Corrupt Practices (RICO) Act).²³⁹ Though these procedures have also been criticized as being exceptional, they may provide sufficient incentive for individuals to bring actions addressing AI-related injuries, including those relating to the use of AI in civil justice settings.

Footnotes