The Ethical Implications of AI in Law for Prospective Paralegals

In nearly every class I teach, artificial intelligence comes up quickly. Students are already using AI tools to summarize readings and draft emails, and the next question is usually the same: How does this translate into legal work?

It is a fair question.

Paralegals often work at the center of drafting, legal research and workflow management, which means they are likely to encounter AI tools and may also be among the people who catch an error before it becomes a larger problem.

The ethical implications of AI in law do not replace existing rules of professional responsibility. Instead, AI ethics adds new complexity to duties that legal professionals already have. Confidentiality, competence, accuracy and supervision still matter, especially because AI-generated work can look polished even when the information is incomplete, misleading or wrong.

Rather than stopping at the general warning that AI cannot replace attorney judgment, this article looks more closely at what bar authorities have said about AI use, why entering case information into a chatbot can create confidentiality concerns, what has happened when attorneys failed to verify AI-generated citations and why checking AI output may become an increasingly important skill for future paralegals.

What do AI tools mean in legal practice?

The ethical considerations around AI use can change depending on the type of tool involved. Not all AI tools work the same way or carry the same level of risk.

Common categories include:

• General-purpose chatbots: These generative AI tools can produce text, summarize information and answer questions across many topics. Because they are not designed specifically for legal practice, they require especially careful review.
• Professional legal AI: These tools are built for legal work and may be connected to licensed, citable sources or established legal research platforms.
• Task-specific legal technology: These tools are designed for narrower legal tasks, such as contract analysis, e-discovery or document review.

For paralegals, understanding the difference matters because the appropriate use, level of human oversight and verification process may vary depending on the tool.

Common ways paralegals may use generative AI

Generative AI may support several everyday legal tasks, including:

• Creating first drafts
• Summarizing documents or information
• Building chronologies
• Extracting data from documents
• Getting initial orientation on an unfamiliar topic before formal legal research begins

Used carefully, these tools may help organize early-stage work. But they do not remove the need to verify information through authoritative legal sources or follow an organization’s policies for approved AI use.

Core ethical obligations still apply to AI use

AI may change the tools used in legal work, but it does not remove the ethical obligations that already guide paralegals and the attorneys who supervise them. Those responsibilities come from both attorney rules of professional conduct and paralegal codes of ethics.

The NALA Code of Ethics and Professional Responsibility is especially useful to revisit with AI in mind.¹

Canon 1 prohibits paralegals from giving legal opinions or taking actions that would assist an attorney in violating professional ethics. Canon 4 allows paralegals to exercise professional judgment, but not to substitute their own legal judgment for an attorney’s. Canon 6 emphasizes maintaining competence through education and training.

Competence, judgment and supervision

Together, these Canons help frame several important responsibilities when using AI:

• Maintain competence: Paralegals need enough technological competence to understand the tools they use, including their limits and potential risks.
• Use professional judgment carefully: AI-generated content should not replace attorney judgment or be treated as final legal analysis.
• Support human oversight: Paralegals work under attorney supervision, and AI use should fit within that same structure of review and accountability.
• Avoid passing along unverified work: Providing an attorney with unchecked AI output that later appears in a filing can create professional conduct concerns.

The technology may be new, but the underlying expectations are familiar: know the limits of your role, understand the tools you use and make sure human review remains part of the process.

What ethics guidance says about AI in the legal profession

An ethics opinion is formal guidance from a bar association explaining how existing rules of professional conduct apply to a specific issue. It is not a statute or court decision, but courts and disciplinary bodies may still look to these opinions as authoritative guidance.

As AI use grows in the legal profession, bar associations have started applying existing ethical rules to questions about competence, supervision, confidentiality and professional judgment.

ABA^® Formal Opinion 512 and technological competence

The American Bar Association’s Formal Opinion 512, issued in July 2024, is one of the central pieces of ethics guidance on generative AI.²

The opinion explains that lawyers do not need to become AI experts, but they should have a reasonable understanding of the capabilities and limits of the specific tools they use. That includes recognizing risks such as:

• Inaccurate AI outputs
• Outdated information or training data
• Biased content
• Difficulty understanding context
• Hallucinations, or information that sounds credible but is invented

The opinion also makes clear that technological competence is not static. As legal technology changes, legal professionals need to continue learning how those tools work and where they may fail.

Most importantly, lawyers cannot rely on AI-generated material without an appropriate level of independent review. AI may help with early research or drafting, but it cannot replace the professional judgment required to provide competent representation.

AI supervision and the Model Rules

Formal Opinion 512 also connects AI use to the ABA Model Rules on supervision.

Model Rule 5.1 addresses a lawyer’s responsibility for other lawyers they supervise. Model Rule 5.3 applies to nonlawyer assistants, including paralegals. Together, these rules mean that AI use by both attorneys and support staff should take place within a clear structure of supervision and accountability.

The opinion says law firm leaders should establish policies for permissible generative AI use and make sure lawyers and nonlawyers receive appropriate training. That training may include secure data handling, privacy, confidentiality and the practical limits of AI systems.

For paralegals, the takeaway is straightforward: You should understand your organization’s AI policies, receive appropriate training and know when AI-assisted work requires additional review.

State bar association guidance on AI use

State bar associations have also issued their own AI ethics guidance.

The Florida Bar’s Ethics Opinion 24-1, for example, compares generative AI to a nonlawyer assistant that must be properly supervised.³ California guidance places strong emphasis on confidentiality and warns against entering confidential client information into AI tools that do not have adequate security protections.

The exact guidance can vary by state, so future paralegals should become familiar with the rules and ethics opinions that apply in the jurisdiction where they work. Because AI technology is evolving quickly, that guidance may also change over the course of a legal career.

Why entering client data into AI can create a confidentiality risk

Client confidentiality covers more than sending a file to the wrong person or sharing a document outside the organization. It can include any information related to a client’s representation, whether or not that information has been labeled sensitive.

That matters when using AI tools. Typing details about a legal matter into a chatbot means transmitting client data to an outside system that may have its own terms of use, data retention practices and security policies. Depending on the tool, that information may be stored, reviewed or used in ways the user does not fully control.

Some AI systems may also create internal confidentiality concerns. Information entered into a shared or self-learning tool could potentially become accessible in ways that conflict with internal privacy rules or screening procedures.

For that reason, safeguarding client confidentiality means understanding how an AI tool handles information before using it for legal work. In some situations, informed consent may also be required before client information is entered into certain AI systems.

Why de-identifying information may not be enough

Removing a client’s name does not always make information anonymous. Details such as dates, jurisdiction, transaction structure or unusual facts may still make a matter identifiable.

That is why partial de-identification should not automatically be treated as a safe solution. A better approach is to use only tools that have been approved by the organization, avoid entering client-specific information when possible and understand the tool’s data retention and privacy policies before using it.

When in doubt, the safest course is to ask a supervising attorney or follow the organization’s established policy for AI use.

What happens when AI-generated legal research is not verified?

AI hallucinations are no longer just a theoretical concern. Courts have now documented a growing number of cases involving fabricated citations and other AI-generated errors in legal documents.

Legal researcher Damien Charlotin maintains a public database tracking legal cases with AI-generated errors, which had identified more than 1,600 worldwide as of June 2026.⁴

A few terms help put these cases in context. A brief is a written document filed with a court that supports an argument with cases, statutes and other legal authorities. Sanctions are penalties a court may impose for improper conduct, such as fines, orders to pay costs or referrals to disciplinary authorities.

Courts also operate in levels: trial courts hear cases first, while appellate courts review those decisions. A decision from a higher court may be binding, while a decision from the same level may only be persuasive.

The cases below show why human oversight and independent verification remain essential when AI-generated outputs enter legal research.

Mata v. Avianca: Fabricated case citations

In Mata v. Avianca, Inc., an attorney representing a man injured on an airline flight used ChatGPT in 2023 to research a response to a motion to dismiss.⁵ The tool produced six cases that appeared to support his argument. None of them existed.

When the court could not locate the cases and asked for copies, the citations still were not independently verified. The attorney later submitted material that included invented case text presented as real authority. The federal trial court in Manhattan found bad faith and imposed a $5,000 sanction on the attorneys and their firm.

The lesson is broader than simply avoiding one particular AI tool. If AI-generated case law appears useful, it still needs to be checked against an authoritative source before it is relied on or included in a filing.

Wadsworth v. Walmart: Internal AI can still get it wrong

Wadsworth v. Walmart, Inc. shows that the risk is not limited to free consumer chatbots.⁶

In 2025, attorneys at a large firm filed a motion citing nine cases. Eight of them did not exist, and one included an invented Westlaw-style reference number that made the citation appear more credible. The drafting attorney had used the firm’s internal AI platform and relied on the results without verifying the cases.

The court imposed $5,000 in sanctions and revoked the drafting attorney’s permission to appear in the case. The case is an important reminder that a professional or internal AI system still requires human oversight. A polished citation or familiar-looking database reference is not proof that the authority is real.

Oregon Court of Appeals: AI errors can reach paralegals indirectly

A 2026 Oregon Court of Appeals case is especially relevant for prospective paralegals because the research problem did not begin with someone intentionally using a chatbot.⁷

The court fined an attorney $10,000 after a brief included 15 fabricated citations and nine quotations that did not appear in any real case. The attorney explained that he had relied on research performed by a paralegal. The paralegal had used Google and encountered AI-generated information in the search results without realizing it.

The court still held the attorney responsible for filing unchecked citations. That reflects an important point about professional responsibility: The attorney who signs and submits the legal document remains responsible for verifying its contents.

For paralegals, though, the case carries another lesson. AI-generated material can appear in ordinary search results and secondary sources even when you never open a chatbot. The most reliable safeguard is to confirm the authority in a primary source, such as the actual statute or published court opinion, before passing it along.

Why AI cannot replace Westlaw^® or Lexis^® for legal research

One common misconception about generative AI is that it can replace established legal research tools such as Westlaw or Lexis. The difference is important.

Westlaw and Lexis provide access to primary law, including statutes and published court opinions, through licensed and verified legal databases. They also include citator systems such as KeyCite^® and Shepard’s^®, which help legal professionals determine whether a case has been reversed, criticized or otherwise affected by later decisions.

General-purpose AI systems work differently. They generate plausible language based on patterns in their training data. They do not function like a legal research database and may produce citations that look complete even when the case does not exist or is no longer good law.

That is why AI can be useful as a starting point in legal research, but it should not serve as the final authority.

A safer process for AI-assisted legal research

A more reliable research process may look like this:

1. Start with an approved AI tool for orientation. Use it to get a general overview of an unfamiliar topic or generate possible search terms.
2. Move to an authoritative legal research tool. Use Westlaw, Lexis or another approved source to locate the actual statutes and cases.
3. Read the primary source. Do not rely only on a summary of a case or statute.
4. Run the citator. Use KeyCite, Shepard’s or another citator to check whether the authority is still valid.
5. Verify the details. Confirm quotations, jurisdiction, court level and current validity before relying on the authority.

If an AI tool mentions a case that cannot be found in an authoritative database, that is a signal to stop rather than assume the citation is accurate. AI can help legal professionals begin the research process, but human oversight and verified legal sources should remain the final check.

AI use in practice: Do’s and don’ts for paralegal students

Responsible AI use in legal practice starts with knowing where these tools can help and where they require extra caution. The goal is not to avoid AI entirely, but to use it in ways that support ethical legal work, client confidentiality and professional conduct.

Do

• Use AI to get a general overview of unfamiliar legal topics before beginning formal research
• Use AI to help organize thoughts, outlines or first drafts
• Verify legal information using authoritative sources such as Westlaw or Lexis
• Ask a supervising attorney when you are unsure whether AI use is appropriate
• Follow your institution’s policies on approved tools and data security

Don’t

• Rely on AI as a final legal authority
• Cite AI-generated cases or statutes without verifying them in a trusted database
• Enter confidential, proprietary or sensitive client information into unapproved AI systems
• Present AI-generated content as legal advice or attorney-reviewed analysis
• Assume an AI tool understands your jurisdiction, organization or specific legal context

These guidelines can help paralegal students use AI for appropriate legal tasks while keeping human review, ethical responsibilities and client protections in place.

Why AI verification is becoming an important paralegal skill

The court cases discussed earlier share one basic failure: AI-generated or AI-influenced information reached a legal document without being fully checked first.

As AI becomes more common in legal practice, that verification work may increasingly become part of a paralegal’s role. Employers may value professionals who can use approved AI tools, understand their limits and catch errors before they reach an attorney, client or court.

That makes verification more than a safety step. It can also be a form of technological competence that prospective paralegals can describe in interviews and demonstrate on the job.

What AI verification looks like in legal work

Verification can include:

• Confirming every citation in an authoritative legal source
• Running a citator such as KeyCite or Shepard’s
• Reading the actual opinion instead of relying on a summary
• Checking quotations against the original source
• Confirming which court issued the decision and whether it applies in the relevant jurisdiction
• Comparing statutory language with the current code
• Keeping records of how AI was used and what was verified when required
• Escalating questionable AI outputs to a supervising attorney

These steps reflect professional responsibility as much as technical skill. As legal technology continues to evolve, paralegals who understand both the tools and the underlying legal research process may be better prepared to spot problems that automated systems miss.

Avoiding the unauthorized practice of law when using AI

The unauthorized practice of law, often called UPL, generally means performing work that only a licensed attorney is allowed to do, such as giving legal advice or offering a legal opinion.

AI can make that boundary easier to cross because it produces detailed, confident answers very quickly. For example, if someone asks whether a contract clause is enforceable, it would be inappropriate for a paralegal to paste the clause into an AI tool and then present the response as a legal conclusion. Even if the answer is framed as “the AI says,” the person receiving it may still treat that information as legal advice.

Paralegals can use AI to support legal assistance, but the tool should not replace attorney review or professional judgment. The attorney remains responsible for legal advice and final work product, while paralegals support that work within the appropriate scope of their role.

Human oversight still matters

Responsible AI use also depends on transparency within the legal team. If an AI tool was used in a way that affected the substance of the work, that use should be disclosed to the supervising attorney when appropriate.

Human oversight matters for another reason: AI systems can reflect bias or gaps in the data they were trained on. Determining whether an output is accurate, fair and appropriate for a specific legal matter still requires professional judgment.

AI may assist with the work, but accountability does not transfer to the tool. Attorneys and paralegals still need to understand how the information was produced, review it carefully and know when an issue should be escalated.

Will AI replace paralegals?

AI is likely to continue changing how some paralegal tasks are performed. Tools can already speed up first-pass document review, summarization, initial drafting and data extraction. But automating parts of the work is not the same as replacing the role itself.

Paralegal work also depends on judgment, context, communication, institutional knowledge and accountability. AI may generate an answer quickly, but it does not fully understand the legal matter, the organization or the people involved. It also cannot take responsibility when something goes wrong.

As AI integration grows across the legal profession, paralegals who understand both the technology and its limits may be better positioned to use it effectively. The underlying research and analytical skills still matter because you cannot verify work you do not understand.

That may be one of the most important lessons for future legal professionals working with evolving technology: AI can make some tasks faster, but the value of human judgment remains.

Building responsible AI skills for a paralegal career

AI can be a useful part of modern legal work, and learning how to use it responsibly can support technological competence. But AI use does not replace professional responsibility, ethical principles or attorney judgment.

For prospective paralegals, the goal is not to compete with AI or avoid it altogether. It is to understand what these tools do well, recognize where they can fail and know how to verify the work before it reaches an attorney, client or court.

Paralegals often need to navigate ethical lines with care. For more on that, check out Legal Ethics for Paralegals: Essential Principles and Real-World Dilemmas.

KeyCite^® is a registered trademark of West Publishing Corporation
Shepard’s^® is a registered trademark of RELX Inc.
Westlaw^® is a registered trademark of West Publishing Corporation
American Bar Association^® is a registered trademark of American Bar Association
ABA^® is a registered trademark of American Bar Association
Lexis^® is a registered trademark of RELX Inc.

¹NALA Code of Ethics and Professional Responsibility: https://nala.org/certification/certification_old/nala-code-ethics-and-professional-responsibility/
²ABA Formal Opinion 512, Generative Artificial Intelligence Tools (July 2024): https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf
³Florida Bar Ethics Opinion 24-1 (Jan. 19, 2024): https://www.floridabar.org/etopinions/opinion-24-1/
⁴AI Hallucination Cases database, Damien Charlotin: https://www.damiencharlotin.com/hallucinations/
⁵Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023)
⁶Wadsworth v. Walmart, Inc., No. 2:23-CV-00118-KHR, 2025 LX 143273 (D. Wyo. Feb. 24, 2025)
⁷Oregon Court of Appeals record fine (March 2026), The Oregonian: https://www.oregonlive.com/pacific-northwest-news/2026/03/oregon-attorney-slapped-with-record-fine-after-citing-case-law-hallucinated-by-ai.html