How GenAI Works
AI has existed for decades, powering tools like internet search engines, cell phones, and spell check. Lawyers have been using AI since at least the 1990s, when LexisNexis and Thomson West integrated AI into their legal research systems. Traditional AI has set tasks that it performs with predefined outcomes. GenAI, however, is able to create new content.
GenAI tools like ChatGPT and Claude analyze patterns in their training data to generate responses, using probabilistic modeling to predict likely outputs. This modeling identifies potential arguments and outcomes but may favor common over contextually accurate outcomes. This adaptability makes GenAI products suitable for a wide range of tasks, but their performance depends on the diversity, size, and quality of their training datasets.
GenAI users must direct the product to complete a task, such as researching a topic or drafting a document, through a prompt. Prompts provide context to let the product know what the user is looking for in an output. Comprehensive prompts yield better outputs. For example, specifying jurisdiction and context when drafting a choice of law clause ensures accuracy. When creating prompts, include the role or persona of the AI (e.g., plaintiff or defense attorney), the tone of the document, the output format, the issue with plenty of context, and the purpose of the task.
The probabilistic nature of GenAI models creates what the industry refers to as hallucinations. Hallucinations may take the form of inaccurate responses fabricated by the GenAI product, such as fake cases, or misinterpreted existing data, such as incorrect holdings of existing cases. For example, in Mata v. Avianca, Inc., attorneys submitted case citations and opinions generated by ChatGPT to a judge. The cases were fictitious, although they often used the names of actual judges and attorneys in the opinions. Similar cases quickly followed, and attorneys have faced sanctions, discipline, and even suspensions from practicing law for their failure to verify the existence and accuracy of the information they submitted to the courts. Litigation hallucinations have been quick to appear, but other hallucinations, such as those in contracts or wills, may go unnoticed for long periods of time due to the nature of the work.
All GenAI products built on this probabilistic model hallucinate at times. There are ways to mitigate the hallucinations, such as by providing detailed prompt instructions or using a layer of additional vetted information that the GenAI product can pull from, called a retrieval-augmented generation (RAG) layer. There is, however, no way to completely eliminate hallucinations at this point. Many legal GenAI products have incorporated a RAG layer of primary law and sometimes legal secondary sources, reducing the hallucinations to the point where an observant attorney should be able to catch and correct these issues. As the Second Circuit said in Park v. Kim, “At the very least, the duties imposed by Rule 11 require that attorneys read, and thereby confirm the existence and validity of, the legal authorities on which they rely.”
Types of Tasks GenAI Can Perform
GenAI excels in a variety of legal and business tasks, making it a versatile tool for modern legal practice. Below is an overview of its primary applications.
Drafting. GenAI tools can generate initial drafts or refined versions of legal documents based on templates from the training material, RAG, or user-provided content and detailed prompts.
Researching. GenAI products aid in identifying relevant legal precedents and condensing large volumes of information. Diligent attorney oversight is required due to the risk of hallucinations.
Brainstorming and organizing. GenAI products can generate ideas and structure arguments, such as outlining legal arguments and documents.
Summarizing documents. GenAI products can summarize lengthy documents. For instance, deposition summaries can be created with a detailed prompt specifying the output format. Other lengthy documents, such as patent applications and proposed legislation, can also be summarized. Outputs should be fact-checked, especially when interpreting legal theories.
Translating. GenAI tools assist with translations but are not certified, requiring review for accuracy, especially with legal terms.
Editing, revising, and polishing. GenAI products can refine drafts for grammar, style, tone, and flow. The tools can also serve as line-by-line editors, perfect for late-night authors.
Data analysis. Attorneys handling e-discovery or large datasets can use GenAI to review metadata, search for patterns, and uncover significant trends and insights.
Learning and training. GenAI products teach attorneys and staff how to use new common technologies or unused features in preexisting technologies, such as how to mail merge from an Excel spreadsheet to a letter in Word.
Nonlegal applications. GenAI products may assist with back-office tasks, such as developing branding and communication strategies, drafting social media posts including text and imagery, generating boilerplate responses for email receipts or out-of-office messages, adjusting the tone of angry emails, and similar tasks. Additionally, many tools now include capabilities to generate images and videos for multimedia presentations or similar work.
Conducting a Needs Assessment
Before adopting a GenAI product, law firms must evaluate their specific needs and challenges. The first step is to identify pain points in current workflows. Are attorneys spending excessive time on document drafting? Does client data have to be entered multiple times in different products? Involving stakeholders such as attorneys, paralegals, and support staff is vital. These individuals can provide insights into daily challenges and suggest features that would improve their productivity. By incorporating this feedback, the firm can ensure that features align directly with the real-world needs of its staff. Together, the team should develop a clear list of criteria to evaluate potential GenAI tools, identifying “must-have” features, “nice-to-have” capabilities, and features to avoid.
Must-have features are those that are essential for the tool to function effectively within the specific needs of the firm. For example, a GenAI product used for legal drafting must have strong security measures to protect client confidentiality and the ability to integrate with existing firm technologies, such as a document management system. Nice-to-have capabilities enhance functionality but are not critical, such as advanced analytics. Features to avoid are those that could introduce unnecessary complexity, such as products requiring specialized IT infrastructure that the firm cannot support.
To ensure consistency and clarity, it is recommended that firms create a detailed rubric based on these criteria. The rubric should include a scoring system, such as a scale of 1 to 5, where 5 indicates that the product fully meets the criterion, and 1 means that the criterion is not present or completely ineffective. For example, in evaluating security, the rubric could assign a score of 5 to a product that offers end-to-end encryption and multifactor authentication. A score of 1 would indicate a product with minimal or no encryption and unclear data protection procedures. This scoring approach allows firms to objectively assess how well each product meets their needs, ensuring a fair evaluation process. When the same rubric is applied to each GenAI product, comparisons may be drawn between products, providing solid documentation and justified decision-making.
In addition to assessing current workflows and identifying firm needs, a list of key questions to ask about the GenAI tools under consideration should be created to remind the product researcher of due diligence needs and ethical obligations. At a minimum, lawyers need to make sure that their ethical obligations to keep their client information confidential are met, which means looking at how the product is using your data, who owns or has a right to use both the data you put into the product and the work created by the product, and how your account and stored data are being protected from security threats. A larger list of issues with further details follows.
Data use. How does the tool collect, store, and use data? Are user inputs being used to train the AI further, which means that your data may be made available to other users as it is now part of the “training material,” and if so, is there an option to disable this? Is the company reviewing prompts and outputs for internal improvements or to ensure compliance with output abuse clauses?
Both input and output questions can be challenging, as the terms of service may be vague about the company’s use of data, and they often incorporate other policies by reference. For instance, Anthropic PBC’s Claude.ai consumer terms of service state that “you retain any right, title, and interest that you have in the Inputs you submit. Subject to your compliance with our Terms, we assign to you all of our right, title, and interest—if any—in Outputs.” In that same clause, it goes on to state: “We may use Materials to provide, maintain, and improve the Services and to develop other products and services. We will not train our models on any Materials that are not publicly available, except in two circumstances: [feedback and acceptable use policy violations].” In February 2025, Anthropic launched the Anthropic Economic Index and published a related paper that examined AI’s effects on the economy and the labor market. The paper looked at the prompts input by free and paid account users to create the dataset. While the millions of prompts used removed the account holders’ information in the anonymization process, it did not change the data used, which potentially contained confidential and sensitive information. The emphasized sentence in the terms above allowed Anthropic to use the information input by its users for this purpose. These terms and usage permissions often change when you have an enterprise-level contract.
The last question above about abuse clauses also caused consternation for several law firms when they missed the clause in one of Microsoft’s AI service terms that stated that user input may be reviewed by Microsoft employees to ensure compliance with its abuse monitoring policy.
Data ownership. Does the company claim any rights to the data you provide? Who owns the output? Does the company retain the right to use your output? Is the output unique to you, or could another user have received it previously and obtained ownership?
As seen in the Claude terms above, many GenAI products release their interests in the data that they receive from you and that they provide to you. However, this does not guarantee that you own the output, as the company can only extend to you the rights it currently has over the material, and it may have already given the rights to that output away. For example, OpenAI’s terms of use state:
Ownership of content. As between you and OpenAI, and to the extent permitted by applicable law, you (a) retain your ownership rights in Input and (b) own the Output. We hereby assign to you all our right, title, and interest, if any, in and to Output.
Similarity of content. Due to the nature of our Services and artificial intelligence generally, output may not be unique and other users may receive similar output from our Services. Our assignment above does not extend to other users’ output or any Third Party Output.
Thus, you may or may not own the output. There is no way to know if you have received unique content that you now own.
While legal documents often copy from one another, and this is generally not seen as an ownership issue, some output may need to be unique and be owned. In these cases, use of generative AI is not recommended.
Security. What protections, such as encryption, are in place for sensitive client information? Where is your information stored? If data is stored outside of the country, would different laws regarding data access apply? What requirements are in place for secure user access, such as multifactor authentication?
Copyright. This is closely related to data ownership. Do you have the authority under copyright laws to use the input you are uploading? Who owns the copyright of the output? How do you make sure that you are not receiving copyrighted information as an output?
Copyright concerns may arise with both inputs and outputs. Most GenAI products require that you have the right and authority to use the documents that you input through their terms of service. So, for instance, if John Smith wants to see more Harry Potter stories and he uploads a digital copy of the books, this would likely breach the terms of service as he does not have the authority or right to upload these copyrighted books. Depending on the GenAI product used, John may also have just injected the copyrighted material into the training material of the AI product, which may then mean that another user receives copyrighted outputs. While it is unlikely that another user would receive a copy of the full text, they may get the description of the delivery of Harry’s letters if they ask the AI to write a story about receiving a letter from Hogwarts. Whereas, if you are working on a case for J.K. Rowling regarding the Harry Potter books, you may have the authority to upload the books, but you would need to refer to the previous section on data use to make sure that it is protected and not shared when uploaded.
Regarding outputs, as seen under data use, the GenAI companies often release their interest in the outputs, but they do not guarantee that the output is unique to you. Several GenAI companies offer indemnification to enterprise accounts for third-party claims of intellectual property infringement. The lower-level accounts are not afforded the same protection. It is the author’s experience that copyrighted material may be provided by the general commercial services when requested; for example, for months, when prompted to “draw a picture of a pokemon eating pizza,” ChatGPT would inevitably create an image of Pikachu eating pizza. The Pokémon issue has since been corrected, but others may still occur.
U.S. copyright law does not currently allow fully AI-created content to be copyrighted, although AI-assisted creations may be copyrighted. The Copyright Office released guidance on the copyrightability of AI-generated and AI-assisted content in January 2025.
Compatibility. Does the tool integrate with existing systems, such as case management software or document automation tools?
Training and RAG data access. What data is available to the GenAI product for prompt responses? Is the data complete? Currently, LexisNexis, Thomson Reuters, and Fastcase/vLex have the most complete collections of primary U.S. law, including judicial opinions, statutes, constitutions, and administrative law. New companies must locate a source of law, such as a public database or government websites, and often do not have the full date range of materials. Understanding the breadth and depth of accessible data is critical for ensuring that the AI can perform tasks accurately and reliably and that your work product has the full range of relevant law.
Vendor reliability. Is the provider an established company or a startup? Does it have experience working with legal professionals, and does it understand the ethical obligations involved? Does it have enough staff to provide needed technical support in the desired time frame?
Updates and expansion. Are there plans to add new features? How frequently will the product be updated? Will the terms of service change with new features and updates? This last question is very important, as it can affect client data and the services you are able to provide. For example, OpenAI’s terms of use give all rights to a user’s output to the user. When OpenAI released its advanced voice mode feature in November 2024, it added a clause to its service terms, which are incorporated by reference into the terms of use. Clause 8 restricts voice use to noncommercial uses and states: “Any rights in Output assigned to you do not include ChatGPT Voice Output,” which could jeopardize client data that was entered via the voice feature and repeated in the output.
Once a needs assessment is complete and vendor questions for due diligence have been established, the firm is in a good position to continue with the process.
Setting a Realistic Budget
Rather than letting the product dictate the budget, it is better to let the budget dictate the product. When planning for the adoption of GenAI tools, it is important to consider both the initial and ongoing costs associated with implementation and establish a budget. Initial costs typically include setup expenses, staff training, and costs for integrating the tool into existing systems. Ongoing costs encompass recurring expenses such as software licenses, hardware upgrades to meet increasing minimum system requirements, software updates, and technical support. These costs can vary significantly depending on the vendor and the level of service required. By planning for both types of expenses early in the planning process, firms can develop a realistic budget for GenAI product integration.
Pricing of legal GenAI products is complicated and evolving. Many firms negotiate contracts specific to their firms for legal services, so the costs may differ from firm to firm. The cost is also further obscured because several of the larger legal research companies have historically required legal service providers and law schools to sign nondisclosure agreements regarding the pricing arrangements of their plans. In addition, companies have varied widely in their charges. For instance, the author was quoted $360 per month to upgrade a bar association–provided Fastcase/vLex account to include Vincent AI. Meanwhile, AI.Law at one point offered to draft a complaint for $49; it has since switched to a subscription model that varies in cost depending on the tools required, ranging between $199 and $699 per month. Steven Lerner looked at pricing for legal GenAI products for Law360 and summed it up as “they’re not cheap,” quoting monthly pricing for tools ranging from $299 to $2,000.
While the pricing is high, it will eventually come down. It also makes the nonlegal GenAI systems attractive to attorneys at $20 per month for individual accounts. While some uses with these lower-cost systems are fine, remember the questions to ask above before using the services with client data or for confidential or sensitive data requirements. Enterprise pricing, where greater protections are available, is not listed on their websites.
Your budget still needs to be the driving force behind GenAI product choice. Going beyond your budget may cause financial issues that can cause greater issues for your firm’s viability.
Creating an Acquisition and Implementation Plan
When creating an acquisition and implementation plan for GenAI tools, it is essential to clearly define roles, establish timelines, and outline support requirements to ensure smooth integration into the firm’s workflows.
Each role within the process plays a vital part in a successful implementation. Roles to define include:
- Research team conducts preliminary investigations and vendor outreach to identify potential tools that meet the firm’s needs.
- Testing team, including representatives from all user groups such as attorneys, paraprofessionals, and interns, evaluates the tools under real-world conditions.
- Decision-makers are responsible for selecting a product and finalizing vendor agreements.
- Integration lead manages workflow integration. Often, a vendor will be able to assist with integration tasks, such as data migration, but it will need a firm contact to coordinate everything.
- Training coordinator ensures that all users receive the knowledge and training to utilize the new tool.
- Evaluators regularly assess post-implementation performance to identify areas for improvement and ensure that the tool continues to meet the firm’s evolving needs.
These key roles should have well-defined tasks and duties in the plan, and the persons holding each role should be identified.
For smaller firms, some of these roles may blur or be eliminated. For instance, if you are a solo firm, you likely do not need a training coordinator unless you also have support staff. Some of these roles may also be outsourced. For instance, many firms hire IT as needed, and you may need to include a consultation with the IT company to make sure that your systems are compatible in the research phase. Training coordinators may also be an area that may be able to be outsourced, but this should be part of the plan.
Timelines are also important to the success of the project. Establishing clear milestones, such as completing research, scheduling demos and trials, making final selections, and implementing training programs, helps keep the project on track. Written timelines also establish realistic expectations for stakeholders and allow for coordination of other large projects that could adversely affect or be affected by the new technology. Depending on the size of the firm, the complexity and importance of the product being integrated, and the stakeholder buy-in, implementation could range from a few weeks to over a year.
Additionally, firms should set specific support guidelines to address potential technical issues promptly. This includes defining the cost for technical support, hours of operation, expected response times, and contact methods such as phone, email, or chat. Establishing these guidelines ensures that all stakeholders know how and when to seek assistance when the GenAI tool does not behave as expected.
By combining detailed planning with diverse input, firms can ensure an efficient and effective implementation process. The plan will be in draft format until the product is selected and all details are entered into the plan. Going out to review live products with the plan started will be helpful to keep you on track and avoid high-pressure sales tactics, like a sense of urgency or a deal if you just sign now.
Understanding the Product Landscape
Navigating the growing market of GenAI tools requires an understanding of the variety of products available, keeping in mind that any list provided is illustrative rather than exhaustive. These tools offer a wide range of legal tasks, from drafting and research to broader administrative and marketing needs. However, the market is dynamic, with new products frequently entering and others exiting without notice. Several primary legal use categories are listed below, along with example GenAI products in each category. These products are not exhaustive but rather representative of these categories.
Research. For research purposes, tools from established legal vendors like Lexis+ AI, Westlaw AI-Assisted Research, and Fastcase/vLex’s Vincent AI stand out. These tools help attorneys summarize case law, locate relevant statutes, and analyze judicial opinions. The way GenAI products function has given rise to new features in some of these systems, such as Vincent AI’s ability to create 50-state surveys on any topic in a few minutes. If the firm has an existing contract for these services, the GenAI features may be added on. If your bar or similar association provides a free research service, such as Fastcase, the service may offer discounts to add its GenAI features to the free account. Before you add the services, make sure that you go through at least an abbreviated form of the process listed here. You still need to do your due diligence; ask questions about data use, data ownership, and security; determine the pricing; test the products; and similar tasks. Adding on to a preexisting service can be more cost-effective, however, as you will already have many of the questions answered and systems in place.
Some of the newer research companies may acknowledge potential risks associated with emerging products, such as lack of comprehensive primary law coverage, limited technical support, or untested reputations, and offer free or more competitively priced services to mitigate these concerns. For instance, Descrybe.ai offers free access to its research platform. It also offers the notable feature of full functionality in both English and Spanish. Midpage.ai offers a free trial and reduced pricing. It also offers a unique research output format in the form of a grid-based result list that allows for quick review of multiple issues.
Drafting. In the realm of legal drafting, GenAI products have a range of drafting abilities. CoCounsel is a drafting and research tool originally created by Casetext, now owned by Thomson Reuters. CoCounsel was under development before ChatGPT was announced to the world in November 2022. Due to its prolonged development, it is a highly refined product. CoCounsel drafts a variety of documents, including correspondence, memos, deposition questions, document summaries, and case timelines. It also assists with contract review and compliance. Its features and document types expand frequently. Lexis+ provides similar features in its newly released LexisNexis Protégé product, which is being billed as a personalized legal AI assistant. In addition to Lexis+ AI, Protégé integrates with Lexis Create+, which is a Microsoft Word add-on.
Spellbook specializes in contract drafting and review, including the creation of a clause library and comparison to industry contracts. Its newest feature allows users to work with multiple documents, connecting and unifying the documents, allowing cross-document updates, and similar tasks. CoCounsel and Lexis+ offer contract drafting tools with similar features as well.
Specialized tasks. Certain tools are designed for specialized practice areas or tasks. For instance, Visalaw.ai GEN provides immigration attorneys with targeted immigration data that is not in other research databases, including law, secondary sources, and document samples. In addition to research, GEN provides drafting capabilities, document review features, and translation tools.
Similarly, litigation preparation can benefit from tools like Clearbrief, CoCounsel, and Protégé, which streamline the process of organizing evidence, preparing arguments, and searching and reviewing case details across multiple documents. Clearbrief enhances the litigation process with hyperlinked fact sections, automatic preparation of exhibits that are cited in your documents, timeline creation, integration with Microsoft Word, and creation of linked documents that can be provided to the courts, regardless of the court’s prior access to the product. One of the most interesting features is that when working in your pleading, Clearbrief can search the rest of your documents to provide factual support for a point that you are making. All three systems can also now analyze your factual documents to create timelines for your complex cases. All three can check citations to make sure that they exist, and Protégé and CoCounsel also connect to the Lexis and Westlaw citators, respectively.
Nonlegal tasks. GenAI products can significantly enhance marketing and administrative tasks. It is now quick and easy to create vibrant websites by describing your vision. Social media or blog posts, complete with images, hashtags, and links, can be generated in seconds. The systems can even provide you with different versions from which you can select. Administrative tasks such as sending reminders or holiday cards can be quickly generated and automated. GenAI can be installed on websites to assist potential clients in completing intake forms, sounding like they are chatting with a human rather than completing a long form. Meetings and phone calls can be automatically transcribed. Even angry emails can be civilized in seconds rather than sleeping on the issues overnight. As with all GenAI content, they will need review, but the time saved in the initial creation may be significant.
There are a wide variety of tools available for media creation, marketing, customer assistance, advising, data analysis, and many other tasks. Google has published a list of 601 real-world examples of business use cases. Examples of tools that may fall into this category include design tools, such as Canva’s Magic Design AI tool and Adobe’s Firefly image generator. General commercial AI systems like ChatGPT and Claude assist in generating content for newsletters, social media, presentations, and client communications. Audio and video editing tools, such as TechSmith’s Audiate and Descript, allow creation and modification of audio and video media via text. If you can think of a task, there is likely a GenAI product to help you. There’s An AI For That tracks AI systems for over 13,000 tasks. As of April 2025, there are over 34,000 AI products being tracked by this site.
One caveat regarding this last category. Most tools are not designed with an attorney’s ethical obligations in mind. Data use and privacy may not be at the level required for use with client information. While some products may let you negotiate terms, many, including OpenAI, do not allow you to do so for individual accounts. Even the paid version of ChatGPT contains potential client data issues, such as the terms of service clause that states: “We may use Content to provide, maintain, develop, and improve our Services, comply with applicable law, enforce our terms and policies, and keep our Services safe.” While you can opt out of data going into the training system, there are many other potential uses permitted in this clause. One of the most important skills for evaluating GenAI products is the ability to read and understand terms of service.
While these tools highlight the diverse capabilities of GenAI, thoroughly evaluating each option aligns firm needs, workflows, and objectives. The goal is to narrow potential tools into a short list for analysis, ensuring that the tools provide the most effective solutions for the firm’s specific requirements.
Engaging with Vendors
Once you have established the universe of products that may meet your needs, prioritize those most aligned with your firm’s specific requirements and conduct further research to determine which ones warrant deeper evaluation or testing within your firm. Reviewing product web pages will assist you in evaluating technical aspects, such as compatibility with existing systems, security features, available tasks, and features. However, vendor outreach is needed to clarify ambiguous details, address specific questions about customization options, and understand the level of support and updates offered.
When engaging with vendors, it is important to schedule demonstrations and discuss key topics, such as licensing agreements, customization options, and data privacy policies, to ensure that the GenAI product meets your firm’s needs. Begin by reviewing the terms and conditions of use, focusing on areas such as licensing agreements, usage limitations, and compliance with legal standards. Next, inquire about possible contract term negotiation. Finally, prioritize discussions on data privacy policies to understand how the vendor handles sensitive information, including storage, security measures, and user data usage. These conversations will provide valuable insights into the product’s capabilities and the vendor’s reliability.
Conducting Trials and Collecting Feedback
Involve a range of users in trials to assess the ease of use and user interface issues. Accuracy and reliability metrics determine the tool’s ability to deliver consistent and precise results. Additionally, trials help gauge how well the product aligns with the predefined criteria created during the needs assessment, ensuring that the chosen solution meets the firm’s unique requirements.
Making a Decision
Once your data has been collected and evaluated using the rubric for each product under consideration, the decision-making team will be able to make an informed choice about which GenAI tool best satisfies the firm’s needs within the predetermined budget. By systematically applying the rubric, the team can ensure that the selection is driven by objective analysis rather than external hype, focusing on the product’s capability to meet specific firm technical and ethical requirements.
Updating and Implementing the Plan
Once the GenAI tool has been selected, the acquisition and implementation plan must be updated to reflect the final choice. This includes incorporating product-specific timelines, updated roles, and responsibilities for deployment; training schedules; and ongoing support needs. Clear communication with stakeholders during this phase ensures that everyone involved understands their part in the implementation process and the timeline for integration and training.
Reevaluating as Needed
Tools should be regularly reevaluated to ensure that they continue to meet the evolving needs of the firm. This process involves assessing the tool’s effectiveness and accuracy, identifying areas for improvement, and determining whether updates or replacements are necessary as new tools and features become available. It is also essential to monitor changes in the terms of service and security policies that may adversely affect the firm and its clients.
Risks of GenAI Product Adoption
GenAI presents both significant risks and rewards for the legal profession, many of which have been discussed throughout this article. Understanding these elements allows legal professionals to use the technology effectively while avoiding potential pitfalls.
The adoption of GenAI tools introduces a spectrum of risks that must be carefully monitored and managed. Ideally, firms will respond to these risks by establishing clear practices and procedures, ensuring alignment with ethical and legal standards.
Confidentiality concerns. Client confidentiality is one of the most significant concerns with GenAI and any cloud technology, as the information entered into these products is placed under the control of a third party. GenAI products frequently require access to sensitive client data to function effectively and provide the best results. Without safeguards in place, such data could be vulnerable to breaches, misuse, or inadvertent sharing with third parties. Data protection policies and practices provide guidance to users on how to ethically and legally navigate these issues.
Accuracy. Another significant risk involves the accuracy of AI-generated outputs. GenAI systems, while powerful, are not immune to errors. Issues such as hallucinations can lead to flawed legal documents or incorrect analyses. These inaccuracies may result in adverse outcomes for clients, such as litigation risks or financial losses, and attorney discipline or sanctions. Attorneys must verify AI-generated content before relying on it in practice.
Data ownership. Data ownership and potential copyright issues also represent concerns in the integration of GenAI into legal workflows. Many GenAI tools are trained on publicly available data, which may include copyrighted material. Using outputs generated from such sources can inadvertently lead to copyright infringements, especially if the AI fails to cite or properly attribute the original sources. Additionally, questions about who owns the input and output generated by the AI remain unresolved in many jurisdictions, creating uncertainty over intellectual property rights and confidentiality obligations.
Judicial and client expectations. The use of GenAI remains controversial among some judges and clients. Before submitting work generated or assisted by GenAI, it is essential to consult ethics opinions, local rules, and court websites to identify any prohibitions or requirements, such as certifications of GenAI use. Duke University’s Responsible AI in Legal Services (RAILS) program provides a resource tracking these rules and orders.
Client preferences regarding GenAI use also vary. Some clients may encourage its use for efficiency and innovation, while others may explicitly prohibit its use in their matters. Firms should disclose their use of GenAI and, where necessary, implement processes to verify client permissions before beginning work. Such procedures ensure that all team members assigned to a matter are aware of and adhere to the client’s expectations regarding AI tools.
Rewards of GenAI Product Adoption
Despite these risks, the rewards of GenAI are undeniable.
Time. By automating repetitive tasks such as document drafting and legal research, GenAI frees up valuable time for attorneys, allowing them to focus on high-value activities like client counseling and strategy development. This shift not only improves efficiency but also enhances the overall quality of legal services. It also offers more opportunities for a better work-life balance.
Cost savings. By streamlining workflows, GenAI reduces the time and resources required for routine tasks, leading to substantial financial benefits. For instance, some immigration law firms have previously spent attorney time generating initial drafts of expert letters for visa applications. Combining letter templates with client background information and a detailed prompt can turn three hours of drafting work into half an hour to draft and review before sending the letter off to the expert for further revision and verification.
Client services. In addition to operational benefits, GenAI enables improved client service. With tools that produce timely and thorough work, firms can respond faster and more fully to client needs. This increased responsiveness fosters stronger client relationships and establishes the firm’s reputation as client-centered and forward-thinking. Together, these rewards make GenAI a tool for growth and innovation in the legal field.
Mitigation Strategies
To maximize the rewards while minimizing risks, there are several strategies that firms may adopt.
Knowledge. Understanding how GenAI technology functions assists with managing expectations regarding output results and creating procedures for GenAI use. GenAI systems rely on complex algorithms trained on large datasets, and their functionality is driven by identifying patterns and probabilities. By understanding these fundamental principles, legal professionals can better evaluate the strengths and limitations of AI tools, such as their susceptibility to errors or hallucinations. This knowledge enables firms to make informed decisions about how to implement these tools responsibly and effectively.
Policies and procedures. Data policies ensure secure, ethical management of client information. These policies should include guidelines for data access, storage, and sharing, with regular updates as technology and regulations evolve. The entire staff should be trained on and have access to these policies and procedures.
Accuracy audits. In addition to data policies and procedures, accuracy should be monitored by auditing AI-generated content to identify and correct inaccuracies before they escalate into larger issues. Such audits should involve a combination of manual review and automated checks to verify the reliability and consistency of AI-generated results. This practice not only enhances confidence in the tool but also reduces the likelihood of errors impacting clients.
Training. Finally, ongoing training for all users is essential to ensure that the technology is utilized effectively and responsibly. Training should cover both the technical aspects of the GenAI tool and the ethical considerations of its use in legal practice. By equipping attorneys and staff with the necessary skills and knowledge, firms can reinforce their commitment to high professional ethical standards while fostering a culture of innovation.
By taking a proactive and informed approach to risks and rewards, firms can harness the potential of GenAI while upholding their professional and ethical obligations.
Conclusion
Generative AI offers significant potential for legal professionals to enhance efficiency, effectiveness, and innovation. However, successful adoption requires thoughtful evaluation, strategic planning, and ongoing oversight. Following this framework, attorneys can harness GenAI while mitigating risks.
