Artificial intelligence is no longer a distant technology issue for courts, lawyers, public agencies, universities, health systems, insurers, employers, legal aid providers, and other Michigan institutions that interact with the civil justice system. It is already shaping how people search for legal information, draft documents, organize evidence, translate complex language, triage disputes, and decide whether to seek help. For Michigan institutions, the central question is not whether AI will affect access to justice, but whether its use will expand meaningful access while preserving fairness, reliability, confidentiality, and public trust. The promise is real: generative AI can help explain legal rights, simplify court procedures, reduce administrative burdens, and support people who cannot afford traditional representation. The danger is equally real: the same tools can produce false legal answers, fabricated citations, biased outputs, privacy violations, unauthorized legal advice, and litigation exposure if institutions deploy them without governance. Michigan’s courts and legal community have already recognized both sides of the issue, including AI’s potential to improve court operations and access to justice, and the need for education, risk controls, human review, and careful implementation. ¹
Michigan’s access-to-justice challenge provides the context for evaluating AI. Many civil legal problems involve housing, family law, debt collection, benefits, employment, consumer issues, probate, and protection from abuse. These matters are often legally consequential but financially unsuitable for full-service representation. A person facing eviction, a custody dispute, a garnishment, or a benefits denial may need urgent, accurate, plain-language guidance before any lawyer is available. Michigan has already invested in self-help resources, court forms, MiFILE, court-based self-help centers, limited-scope representation, and statewide access-to-justice planning. The Michigan Justice for All effort reflects the broader goal that residents should be able to understand, enter, and navigate the civil justice system regardless of income, education, geography, disability, language, or representation status. ² AI tools should be judged against that reality. They should not be compared to a perfect lawyer available to everyone at no cost, because that is not the current system. The fair comparison is whether a properly limited, tested, and supervised AI tool can help more people obtain reliable information, complete routine forms, identify deadlines, prepare for hearings, and reach appropriate human assistance sooner than they otherwise would.
AI can expand court access most immediately by improving public-facing legal information. Many people do not know whether their issue is legal, where to file, what form to use, what deadline applies, or what words in a court notice mean. A well-designed AI assistant trained or constrained around approved Michigan court materials could help users understand the difference between legal information and legal advice, ask clarifying questions, point users to official forms, and explain procedural steps in plain language. This kind of tool could be especially useful for self-represented litigants who struggle with dense terminology such as “default,” “service of process,” “motion,” “proof of service,” “answer,” “summons,” “show cause,” “adjournment,” or “judgment.” It could also support courthouse staff by answering repetitive procedural questions while preserving the staff’s neutrality. The key is that the tool must be designed as a navigator, not a substitute judge, lawyer, or clerk. It must explain limits, avoid strategic advice unless supervised by authorized legal professionals, and direct users to legal aid, lawyer referral, self-help centers, or emergency resources where appropriate.
AI may also improve the accessibility of legal language. Court documents are often written for legally trained readers, not ordinary people in crisis. A generative AI tool can summarize a notice, translate formal phrasing into plain English, identify action items, and help a user prepare questions for a lawyer or court clerk. For Michigan institutions serving the public, this can matter beyond courthouses. Universities responding to student conduct proceedings, hospitals handling guardianship or treatment-related petitions, municipalities processing ordinance matters, housing authorities administering notices, and nonprofits helping clients with benefits or family issues all produce documents that people may not understand. AI can help rewrite public-facing materials at lower reading levels, create multilingual explanations, and generate accessible versions for individuals with disabilities. Done properly, this use supports procedural fairness because people are more likely to comply with rules they can understand. Done poorly, it risks oversimplifying legal consequences, mistranslating key terms, or creating misleading summaries that a user mistakes for the controlling document.
Another opportunity lies in form completion and document assembly. Michigan already has a strong foundation in online self-help forms and court-approved forms. AI can build on that infrastructure by helping users identify which forms may fit their situation, explaining what information belongs in each field, flagging missing information, and warning when a case type may require immediate legal advice. The most responsible use would not be an open-ended chatbot that invents legal documents from scratch. It would be a guided tool tied to approved templates, jurisdiction-specific instructions, and clear review checkpoints. For example, an AI-enabled form assistant could help a tenant distinguish between a repair issue, a notice to quit, an eviction complaint, and a money judgment concern. It could then route the user to approved resources and explain what information is needed. In family law, it could help users understand the difference between custody, parenting time, child support, and divorce forms without recommending litigation strategy. In debt collection, it could help identify whether a person has received a complaint, a default, a garnishment, or a notice of rights. This is access to justice in practical terms: fewer abandoned claims, fewer defaults caused by confusion, and better-prepared litigants.
Court administration is another area where AI may deliver benefits without directly affecting adjudication. Courts and related institutions process enormous volumes of documents, notices, transcripts, emails, docket entries, and public inquiries. AI can assist with summarizing long filings, classifying incoming documents, routing questions to the correct department, generating draft notices, identifying incomplete submissions, and improving internal knowledge management. The Michigan Judicial Council’s work on generative AI recognized potential uses in court operations, internal processes, education, training, risk awareness, access to justice, and the rule of law. ¹ These operational uses may be lower risk than AI tools that directly recommend outcomes, provided they remain subject to human review. A court might use AI to summarize a file for staff preparation, but not to decide credibility. A clerk’s office might use AI to draft a response to a procedural question, but a trained employee should review the response before it is sent. A legal aid organization might use AI to triage intake, but the triage system should not quietly screen out people with meritorious claims because their facts are unusual or their language does not match the model’s training data.
AI also has potential in language access and disability access. Michigan residents include people with limited English proficiency, people with low literacy, people with visual or hearing impairments, people with cognitive disabilities, and people who cannot physically reach a courthouse during business hours. AI-driven translation, speech-to-text, text-to-speech, plain-language conversion, and guided question systems may help institutions meet users where they are. However, this is one of the areas where optimism must be matched with caution. Legal translation is not ordinary translation. A mistranslated court order, waiver, plea-related advisement, custody instruction, or benefits deadline can have serious consequences. Institutions should avoid using AI translation as an unsupervised replacement for qualified interpreters where legal rights are at stake. The better approach is to use AI for preliminary access, public education, and administrative convenience while preserving certified interpretation and human review for official proceedings, legally operative documents, and high-stakes communications.
The litigation risks begin with accuracy. Generative AI systems can produce confident but false statements, including fabricated legal authorities, nonexistent rules, wrong deadlines, and misleading summaries. In litigation, that risk can become sanctions exposure, malpractice exposure, consumer protection exposure, due process challenges, or reputational harm. Courts and lawyers across the country became alert to this issue after filings appeared with AI-generated cases that did not exist. Michigan’s judicial AI workgroup likewise emphasized hallucinations, deepfakes, unreliable outputs, and the need for human review and verification. ¹ For Michigan institutions, this means no AI output should be treated as authoritative merely because it is fluent. A chatbot answer about a filing deadline, a generated summary of a court order, or an AI-created draft affidavit must be checked against governing law, court rules, official forms, and the actual record. Institutions should assume that every AI tool can be wrong, especially when asked about law, facts, citations, procedural deadlines, or individualized legal consequences.
The second major risk is unauthorized practice of law and improper legal advice. Access-to-justice tools often walk a fine line between providing information and advising someone what they should do. A public website may properly explain what an answer to a complaint is, but it may create risk if it tells a user that they should deny a particular allegation, assert a specific defense, or take a settlement position based on individualized facts. The risk is heightened when users believe an institutional AI assistant speaks with legal authority. Michigan legal institutions should therefore define the tool’s role before deployment. Is it giving general legal information, helping complete approved forms, conducting intake for a licensed legal provider, supporting a lawyer’s work, or providing internal administrative assistance? Each role has different risk. A legal aid organization may be able to use AI under lawyer supervision in ways that a court or university help desk cannot. A court must preserve neutrality. A law firm must preserve professional judgment. A public agency must avoid misleading citizens about rights against the agency itself.
Confidentiality and privacy create another serious litigation risk. AI systems often require users to enter facts, names, dates, documents, or sensitive narratives. In the justice context, that information may involve domestic violence, immigration status, child custody, disability, medical treatment, criminal history, financial distress, housing instability, or protected personal identifying information. The State Bar of Michigan’s AI guidance warns that lawyers must understand the benefits and limitations of AI tools, protect client information, use professional judgment, and consider whether client consent is required when confidential information may be exposed. ³ The same principle applies institutionally even outside the attorney-client relationship. Michigan institutions should not allow employees to paste pleadings, medical records, student files, personnel documents, protected health information, sealed records, juvenile information, or confidential settlement materials into public AI tools. Vendor contracts must address data retention, training use, encryption, access controls, audit logs, breach notification, subcontractors, jurisdiction of stored data, and deletion rights. Without those controls, an efficiency tool can become a discovery exhibit.
Bias and unequal impact are also central concerns. AI systems are shaped by training data, design choices, prompts, evaluation metrics, and user behavior. If historical legal data reflects unequal treatment, an AI system trained on that data may reproduce or amplify those patterns. Risk assessment tools, triage systems, settlement prediction models, fraud detection tools, and eligibility screeners can all produce disparate impacts if not tested. In Michigan, institutions serving diverse urban, rural, tribal, immigrant, low-income, disabled, and elderly populations should be particularly careful before using AI to rank urgency, predict outcomes, identify “high-risk” users, or allocate scarce services. A tool that works well for a represented commercial party may fail a self-represented parent writing in plain language. A model may misunderstand dialect, trauma-affected narratives, translated text, or incomplete facts. If an AI tool quietly directs some users away from remedies, delays assistance, or misclassifies claims, the institution may face litigation alleging discrimination, denial of meaningful access, arbitrary decision-making, or failure to accommodate.
Due process is a related but distinct concern. AI may be useful in administration, but adjudication requires transparency, accountability, and the opportunity to be heard. A person affected by a court-related decision should be able to understand the basis for that decision and challenge adverse information. If an institution relies on an AI-generated risk score, credibility flag, eligibility prediction, or document classification without explanation, the affected person may not know how to contest it. Courts in particular must avoid the appearance that a machine is deciding legal rights. Even when AI is only assisting staff, institutions should preserve a record of what was used, who reviewed it, what data was considered, and what human decision was ultimately made. The more consequential the decision, the stronger the need for disclosure, reviewability, and procedural safeguards. Public trust depends not only on accurate outcomes but also on a process people perceive as fair, human, and accountable.
Evidence risks are growing as well. Deepfakes, synthetic audio, altered images, AI-generated documents, and fabricated screenshots can affect litigation in family law, employment, criminal proceedings, protection orders, insurance disputes, school discipline, and business cases. Institutions must prepare for both sides of this problem. They may need tools to detect manipulated evidence, but they must also avoid overclaiming what detection tools can prove. A detector’s confidence score is not a substitute for authentication under evidence rules, chain of custody, metadata analysis, forensic review, witness testimony, and judicial gatekeeping. Michigan institutions should expect discovery disputes about whether AI was used to create, alter, summarize, search, or select evidence. They should also expect questions about litigation holds, preservation of prompts and outputs, model logs, and whether AI-assisted review missed relevant documents. The safest posture is to treat AI use in evidence handling as a documented process, not an invisible shortcut.
Vendor risk deserves special attention because many institutions will not build AI tools themselves. They will buy subscriptions, integrate third-party platforms, or use AI features embedded in existing software. That creates contract and procurement issues. Institutions should ask whether the vendor uses customer data to train models, whether outputs are indemnified, whether the system is auditable, whether it supports role-based access, whether it complies with relevant privacy laws, whether it can segregate confidential information, and whether it permits meaningful human oversight. A vendor’s marketing claim that a tool is “secure,” “legal-grade,” or “court-ready” should not replace due diligence. The National Institute of Standards and Technology’s AI Risk Management Framework provides a helpful governance model because it treats AI risk as an organizational process involving mapping, measuring, managing, and governing AI systems.⁵ For Michigan institutions, that means AI procurement should involve legal, IT, cybersecurity, records management, compliance, accessibility, procurement, and program staff rather than a single enthusiastic department.
Law firms and legal departments face their own professional responsibility risks. AI may reduce the time needed to draft correspondence, summarize discovery, review contracts, conduct research, and prepare first drafts. But faster work does not eliminate the duties of competence, confidentiality, communication, supervision, candor, and reasonable fees. The ABA’s Formal Opinion 512 makes clear that lawyers using generative AI must consider existing ethical duties, including competent representation, protection of client information, communication, supervision of employees and agents, meritorious claims, candor to tribunals, and reasonable fees.⁴ Michigan lawyers should view that national guidance alongside Michigan-specific ethics guidance. A lawyer cannot delegate judgment to an AI tool, cannot bill unreasonably for time saved by automation, and cannot submit AI-generated work without verifying its accuracy. For institutional clients, outside counsel guidelines should address whether counsel may use AI, what tools are approved, whether confidential information may be entered, how AI-assisted work will be reviewed, and how AI-related costs will be billed.
Public agencies and regulated institutions must also consider transparency and records obligations. If an AI tool helps draft public communications, summarize complaints, evaluate applications, or classify enforcement matters, the institution should consider whether prompts, outputs, logs, and model settings are records that must be retained. Even when not legally required, documentation may be essential to defend against later claims. A litigant may ask whether an adverse decision was influenced by AI, what data was used, whether the system was validated, whether similarly situated people were treated differently, and whether the tool was tested for bias. Institutions that cannot answer those questions may appear careless even if the outcome was correct. A defensible AI program should therefore produce an audit trail proportionate to the risk of the use case.
For Michigan courts and court-adjacent institutions, a phased approach is wiser than a dramatic launch. The best first uses are usually internal, low-risk, and reversible: summarizing nonconfidential training materials, drafting internal meeting notes, improving website explanations, organizing frequently asked questions, and helping staff locate approved resources. Moderate-risk uses, such as public chatbots, form assistants, intake triage, translation support, and document summarization, require stronger safeguards, including testing, disclaimers, escalation pathways, data controls, and periodic review. High-risk uses, such as outcome prediction, risk scoring, credibility assessment, automated eligibility denial, or decision recommendations, require the highest scrutiny and may be inappropriate in many court-access contexts. The Michigan Judicial Council’s report supports this cautious posture by emphasizing training, pilot programs, risk awareness, ethics, professional responsibility, and implementation consistent with court objectives. ¹
Training is essential because AI risk is often created by ordinary users, not only by software design. Employees may paste confidential information into a tool because it is convenient. Staff may rely on an AI summary because it sounds polished. Lawyers may fail to check citations because the draft appears professional. Administrators may approve a vendor without understanding data retention. Judges and court staff may encounter AI-generated filings without knowing how to evaluate them. Michigan institutions should train users to understand what AI is good at, where it fails, what information must never be entered, when outputs must be verified, and how to escalate concerns. Training should also make clear that AI is not a neutral oracle. It is a tool that produces outputs based on data, design, probability, and instructions.
A sound AI policy should be written in practical language. It should define approved tools, prohibited uses, confidential-data restrictions, review requirements, citation-checking obligations, recordkeeping rules, vendor approval processes, and consequences for misuse. It should distinguish between public tools and enterprise tools. It should require human review for external communications and legal work. It should prohibit unsupervised AI legal advice to the public. It should require accessibility and bias testing for public-facing systems. It should establish a process for reporting erroneous or harmful outputs. Most importantly, it should assign responsibility. An AI policy that does not identify who owns risk, who approves tools, who monitors performance, and who can suspend a tool is unlikely to protect the institution when problems arise.
Institutions should also involve the communities they serve. Access-to-justice technology often fails when designed around institutional convenience rather than user needs. A chatbot that works for lawyers may confuse self-represented litigants. A form assistant that assumes stable internet access may not work for rural users, elderly users, or people relying on phones. A translation tool may perform unevenly across languages. A triage tool may miss urgent domestic violence issues if users do not describe their problem in expected terms. User testing with self-represented litigants, legal aid providers, court staff, librarians, disability advocates, language-access professionals, and community organizations can reveal risks before launch. Michigan’s Justice for All work already recognizes the importance of welcoming, understandable, collaborative, adaptive, and trusted systems. ² AI tools should be measured by those values, not only by speed or cost savings.
The best use of AI in court access may be as a bridge between people and human help. AI can help someone understand the first step, gather documents, identify the right courthouse, prepare a timeline, generate questions, or recognize that a matter is urgent. It can help legal aid screen more people and help lawyers spend less time on repetitive drafting. It can help courts communicate more clearly and process information more efficiently. But it should not become a wall that keeps people away from clerks, judges, lawyers, interpreters, advocates, or hearings. If AI becomes a substitute for human access, it may reproduce the very justice gap it was meant to close. If it becomes a bridge, it can help Michigan institutions use scarce human expertise where it matters most.
The litigation risk for Michigan institutions is therefore manageable but not optional. The institutions most likely to benefit from AI are those that treat it as a governed operational capability rather than a novelty. They will identify use cases, classify risk, select vendors carefully, protect data, test outputs, train staff, document decisions, preserve human oversight, and revise policies as the technology changes. The institutions most likely to face claims are those that allow unsupervised adoption, rely on public tools for confidential work, deploy chatbots without legal boundaries, use AI scores without explainability, or fail to correct known errors. In litigation, the question will often be simple: Was the institution’s use of AI reasonable under the circumstances? A documented governance program makes that question much easier to answer.
AI will not solve Michigan’s access-to-justice crisis by itself. It cannot replace adequate funding, legal aid, interpreters, self-help centers, simplified rules, fair procedures, or judicial discretion. It cannot create trust where institutions are opaque. It cannot make biased data fair merely by processing it faster. But AI can help Michigan institutions make legal systems more understandable, responsive, and navigable if it is deployed with humility and discipline. The opportunity is to use technology to reduce confusion, delay, and administrative burden. The litigation risk is that institutions will confuse fluency with accuracy, automation with judgment, and access with abandonment. The right path is neither rejection nor blind adoption. It is careful, human-centered implementation rooted in Michigan’s access-to-justice goals, professional responsibility, risk management, and public trust.
Contact Tishkoff
Tishkoff PLC specializes in business law and litigation. For inquiries, contact us at www.tish.law/contact/. & check out Tishkoff PLC’s Website (www.Tish.Law/), eBooks (www.Tish.Law/e-books), Blogs (www.Tish.Law/blog) and References (www.Tish.Law/resources).
References and Footnoted Sources
- Michigan Judicial Council, “Generative AI and the Courts Workgroup Report and Recommendations,” October 2024. The report identifies generative AI and the courts as a strategic initiative, discusses training, court operations, access to justice, ethics, risk awareness, hallucinations, deepfakes, and human oversight. www.courts.michigan.gov/4aec3b/siteassets/committees%2C-boards-special-initiatves/michigan-judicial-council/2024-genai-wg-report.pdf
- Hon. Brian K. Zahra and Angela Tripp, “Justice for All: Commission Embraces Goal of 100% Access to Civil Justice System,” Michigan Bar Journal, November 2021. The article describes Michigan’s Justice for All work, Michigan Legal Help, statewide self-help tools, MiFILE integration, self-help centers, and the goal of 100% access to the civil justice system. https://www.michbar.org/journal/Details/Justice-for-All-Commission-Embraces-Goal-of-100-Access-to-Civil-Justice-System?ArticleID=4267
- State Bar of Michigan, “Artificial Intelligence FAQs,” last updated November 2024, and State Bar of Michigan, “Transforming the Legal Landscape in the Age of AI,” June 2025. These sources address Michigan lawyers’ technology competence, confidentiality, client communication, professional judgment, billing issues, unauthorized practice concerns, and access-to-justice implications. https://www.michbar.org/opinions/ethics/AIFAQs
- American Bar Association Standing Committee on Ethics and Professional Responsibility, “Formal Opinion 512: Generative Artificial Intelligence Tools,” July 29, 2024. The opinion discusses lawyers’ duties of competence, confidentiality, communication, supervision, candor, meritorious claims, and reasonable fees when using generative AI. https://www.americanbar.org/news/abanews/aba-news-archives/2024/07/aba-issues-first-ethics-guidance-ai-tools/
- National Institute of Standards and Technology, “Artificial Intelligence Risk Management Framework,” including AI RMF 1.0 and the Generative AI Profile, NIST-AI-600-1. The framework provides a governance and risk-management structure for identifying, measuring, managing, and reducing AI risks. https://www.nist.gov/itl/ai-risk-management-framework