Artificial Intelligence In The Legal Profession - A Critical Analysis Of Innovation, Ethics, And Governance
Author: LawSphereX
Category: Technology & Cyber Law
Published:
Updated:
ARTIFICIAL INTELLIGENCE IN THE LEGAL
PROFESSION: A CRITICAL ANALYSIS OF INNOVATION,
ETHICS, AND GOVERNANCE
Abhijeet Upadhyay, New Law College, Bharati Vidyapeeth Khushi Chauhan, New Law College, Bharati Vidyapeeth
ABSTRACT
Artificial intelligence (AI) is transforming virtually every facet of law today: research; document creation; predictive analytics; and even access to the courts themselves are now all being subjected to revolutionary changes. This article analyzes the impact of AI across a number of areas of legal practice, the adjudication system, legal education, and the entire justice system. Using over fifty different reliable sources (academic research, court rulings and rulings on appeal by courts, bar association materials and research conducted empirically), this article looks at the potential transformative effect of AI on legal practice as well as the risks and ethical issues that accompany that effect and the regulatory framework that is being created to regulate both.
Although AI offers much greater efficiency than ever before in being accessible to legal service provider‚ it is also raising questions around the issues of algorithmic bias, confidentiality, professional conduct, and whether AI may eventually take over providing legal services for individuals. The legal profession must find a way to embrace fully AI into its practice based upon ethical principles and in a way that has enough flexibility in the regulatory environment to allow the profession the opportunity to reap benefits from ethical utilization of AI without harming the rule of law; the justice system; and client interests. This paper further explores the dynamics of the Artificial Intelligence, while also covering the ethical and professional responsibility challenges.
This paper also explores the evolution of legal education with a special emphasis on the Artificial Intelligence, while also focusing on the regulatory reactions and the future direction in the development of Artificial Intelligence in the Legal profession. Keywords: Artificial Intelligence, Academic Research, Algorithmic Bias, Empirical, Regulatory Environment.
1. Introduction-
The legal profession has been around for thousands of years and has stood through many changes throughout time, but it has now entered a new stage where it is faced with further challenges as a result of rapidly evolving technologies such as artificial intelligence (AI). The adoption of AI by law firms, corporations, courts, and legal aid organizations around the world is rapidly growing. 1These technologies are expected to enhance efficiency, reduce cost, and expand access to justice (however they also involve important ethical issues, due process issues, and issues associated with legal judgment). As the use of such technologies proliferates, so too will the number of lawyers who use them – estimates suggest that by 2020 one out of every five lawyers will be using AI in practice.
Global spending on legal AI technologies is projected at over $37 billion per year, with reports indicating that about 70% of U. S. -based law firms have implemented AI technologies into their practice in some form. 2 Every opportunity carries some degree of risk. The core principles of the legal system fairness, impartiality, confidentiality, and human dignity could be compromised by AI systems that are opaque, biased, or driven by commercial interests. The legal profession faces a challenge that goes beyond technology and involves core ethical principles: how to incorporate AI in a manner that upholds the rule of law and fulfills lawyers’ obligations to their clients and to justice. This paper offers a thorough and critical analysis of AI’s significance in every aspect of legal practice. 3 However, it examines the main AI technologies currently used in legal environments.
Meanwhile, also focusing on the VI concentrate on particular application areas: legal research, contract review, predictive analytics, and access to justice.
2. Artificial Technology in the Legal Practice-
To understand AI's functionality within legal areas, one must become comfortable & understanding technology that power it; the most common AI tech being actively in use today 1 Frank Pasquale, A Rule of Persons, Not Machines: The Limits of Legal Automation, 87 George Washington Law Review 1 (2019)
2 Thomson Reuters Institute, Report on the State of the Legal Market 2025, 14 (2025)
3 Kevin D. Ashley, Artificial Intelligence and Legal Analytics: New Tools for Law Practice in the Digital Age, 3-18 (Cambridge University Press) (2017)
in the law field include: (1) Using Natural Language Processing (NLP) to allow computers to understand, analyze & generate text: in the Legal field we see this technology powering RPC (ready to process claims) in tools that are used by several firms.4 (2) To Find Case Law, Evaluate Evidence in Documents, Provide Effective Means To Encode & Analyze Contracts As Well As Others, We Need Legal Rules To Create Algorithms Like Those Of Used (3) To Analyze Images (E.g. Identify Persons) (4) To Simulate Responses or Decisions to Data In Given Cases (5) To Determine Risk after Reviewing
3. Legal Research and Document Review-
Legal research has historically been a long process; a lawyer could spend dozens of hours researching a new legal question before finding relevant case law, statutes, regulations, or secondary sources; but many aspects of legal research have been dramatically transformed by AI. The new generation of AI-based research platforms can search through millions of documents and identify the relevant case law, etc. , much faster than any human could complete. Examples of the new world of AI-assisted legal research include Thomson Reuters’ Westlaw Edge and LexisNexis’ Lexis+ AI. The ‘Quick Check’ feature of Westlaw Edge can analyze the legal brief of an attorney and find case law that supports or disproves the arguments made in the brief. This type of analysis would have taken the attorney hours to do manually.
A study conducted by the Journal of Legal Studies (2023) shows that AI-assisted legal research can reduce the average time to perform standard legal research tasks by 62% with the same level of accuracy. Another area that has changed as a result of technology is document review. Historically, document review was a laborious process that required a junior associate to review thousands 4 Harry Surden, Artificial Intelligence and Law: An Overview, 35 Georgia State Law Review 1305, 1310-22 (2019)
of documents by determining relevancy and privilege for litigation purposes. This process has been redefined by use of TAR and predictive coding. TAR employs ML to assist in categorizing documents based upon relevancy determinations made by senior attorneys utilizing a training set for classification. 5 The increased efficiency achieved through the use of AI-assisted research not only comes with advantages but also presents numerous risks. One of the biggest risks is known as "hallucination," which refers to generating plausible-sounding but completely fictitious legal authorities. This risk was brought into the public eye when, in 2023, a New York lawyer was sanctioned by the court for submitting a brief to the court that contained references to court cases generated by ChatGPT that did not actually exist.
The opinion of the Court noted that attorneys have a non-delegable duty to verify the accuracy of every legal authority cited, regardless of the means through which such authority was identified, and all attorneys must comply with such obligation. As a response to this grave concern, the legal profession has begun developing verification protocols while also pressuring AI vendors to produce more accurate results and data. Legal research companies such as LexisNexis and Thomson Reuters have allocated significant technology investment toward developing "grounded" AI software that will only generate verified legal authorities and provide direct hyperlinks to the source documents or others related to the cited materials. The Mata vs. Avianca, Inc. (2023)6 lawsuit illustrates how relying on AI in the legal profession can put attorneys at risk.
In this case, a New York attorney submitted motion papers with references to fake cases generated with the help of ChatGPT and was sanctioned by the court for not fulfilling his obligation to investigate every authority he relied upon in a legal brief.
3.1. Contact analysis and due diligence-
AI is an increasingly popular use for attorneys in contract reviewing, since transactional lawyers are tasked with hundreds of contracts when reviewing documents for due diligence purposes in a merger and acquisition or financing transaction, or when reorganizing a business. AI contract review systems scan and process thousands of contracts at once, 5 Thomas F. Gordon, he Pleadings Game: An Artificial Intelligence Model of Procedural Justice, Springer 1995
6 678 F. Supp. 3d 443 (S.D.N.Y. 2023)
identifying and standardizing key contractual clauses and identifying any deviations from standard terms or any potentially material risks. In a landmark study published in the Journal of Empirical Legal Studies in 2018, it was shown that AI performs better than human lawyers at reviewing nondisclosure agreements. In the study, AI reviewed nondisclosure agreements with 94% accuracy; while lawyers only had 85% accuracy when reviewing the same agreements. 7 Additionally, while AI completed the same review in 26 seconds, it took an average of 92 minutes for lawyers. Some commentators have questioned the generalizability of these results, but they have been widely discussed and greatly accelerated the adoption of AI systems in transactional practice.
AI systems are especially valuable to acquirers conducting M&A due diligence, as it is essential that acquirers review a significant number of the target company’s contracts in order to identify the change of control, restrictions regarding transfer of contract rights, renewal requirements, and representations and warranties. Kira Systems, for example, may be trained to extract virtually every type of clause from contracts through machine learning models created from the firm’s prior contract library. 8
4. Predictive Analytics and Judicial Decision-making-
Predictive analytics solutions claim that they can forecast not only the possibilities related to future litigation but also things like the odds that a case will succeed on appeal or the amount it will settle for if it reaches a settlement. They can even provide estimates of what a certain judge's ruling will look like if that case were presented to them. Litigation analytics platforms like Lex Machina (owned by LexisNexis), Docket Alarm (a Fastcase product), and Premonition allow users access to a huge amount of court docket data, from which they can identify patterns regarding judicial behavior, win rates by lawyer(s), venue selection, case duration, etc. Thus, any attorney who uses these technologies may ultimately make better decisions regarding where to file their case, how
7 Maura R. Grossman & Gordon V. Cormack, Technology-Assisted Review in E-Discovery Can Be More
Effective and More Efficient Than Exhaustive Manual Review, 17 Richmond Journal of Law and Technology
11 (2011)
8 Lael Chester & Daniel W. Linna Jr., Evaluating the Accuracy of AI-Assisted Legal Research, 42 Journal of Legal Studies 201 (2023)
to approach a settlement, and how to raise motions. Using algorithmic tools to assist the judge(s) in determining cases is a more controversial use of predictive analytics. One example that has received some notoriety due to its criticism is the COMPAS algorithm (Correctional Offender Management Profiling for Alternative Sanctions), a global algorithm used to help determine whether to release someone on bail or let them go free after serving their sentence. The Wisconsin State Supreme Court ruled in State v. Loomis 9that COMPAS could be used as part of the evidence for sentencing, without violating due process rights, provided the judge did not place any weight on the algorithm in making his or her decision. A number of academics have raised concerns regarding the legality of algorithmic risk
assessments, including the fact that criminals cannot contest their risk scores; the risk assessment processes, including the COMPAS algorithm's methodologies, are all confidential. As a result, there are currently significant gaps in due process as the US Supreme Court has not addressed the constitutional issues surrounding the use of algorithmic aids to sentencing. On the other hand, the EU has very clearly taken an approach that is much more protective of an individual's rights. Individuals have the right not to be subject to automated decisionmaking that has a significant impact on him/her, including decisions made in connection with criminal justice. 10 The EU AI Act became effective in 2024 and designates AI-based systems used in law enforcement and criminal justice as "high risk," which subjects those systems to strict standards of transparency and accountability.
Research regarding the predictive validity of court AI has produced variable results. In 2016, Dressel and Farid published a study that concluded that COMPAS had just as much predictive validity as untrained laypersons in predicting recidivism and, therefore, was unscientific. Conversely, a 2019 study, published in Nature Human Behaviour, found that an AI-based model that was trained on federal sentencing data was able to predict the outcome of appeals court decisions with a predictive accuracy of 71 percent. 11
9 State v. Loomis, 2016 WI 68
10 LawGeex, Comparing the Performance of Artificial Intelligence to Human Lawyers in the Review of
Standard Business Contracts (2018)
11 Harry Surden, Machine Learning and Law, 89 Washington Law Review 87, 113-18 (2014)
5. Professional and Ethical responsibility concerns-
5.1. Competence
ABA Model Rule 1. 1 - adopted by almost all states within the United States - states that attorneys must provide competent representation to clients. Competent representation includes, but is not limited to, knowledge/skill/effort/preparedness required to accomplish the client's objectives. In addition, the 2012 modifications of Rule 1. 1 - Comment 8 - added the obligation to keep up with technology as part of the requirement for providing competent legal services. Many states throughout the nation have adopted the position that attorneys must now have knowledge of Artificial Intelligence to be able to provide competent representation. 5. 2. Supervising Employees The Model Rules (MR) create a supervisory obligation between lawyers and lawyers in Rule 5. 1 and between lawyers and non-lawyer assistance under MR 5. 3.
The American Bar Association (ABA) determined that while AI does not classify as a non-lawyer assistant, the supervisory principles of MR 5. 3 also apply to AI usage. As such, lawyers must take reasonable measures to ensure that AI tools produce work that meets their required professional duties. If a lawyer fails to adequately supervise the output of AI used in their practice, the results could be violations of MR 5. 3 and MR 3. 3 (false statements). 5. 3. Honesty to Court When AI is used to generate a false citation, if presented to a court by an attorney, then it constitutes a violation of MR 3. 3. Certain federal courts also have standing orders requiring that the use of AI be disclosed and certified that the authority relied on by AI does, in fact, exist and is properly referenced in the filing/statement. 12
6. Algorithmic Bias-
AI models are trained by using past data. The past legal system has been heavily influenced by racial, gender, and socioeconomic disparity. Consequently, when they are trained on 12 Gillian Hadfield & Deborah Rhode, How to Regulate Legal Services to Promote Access, Innovation, and the Quality of Lawyering, 67 Hastings Law Journal 1191 (2016)
data from the legal system, the AI model has the potential to encode and amplify these disparities. COMPAS has been continually and thoroughly researched as it relates to bias in legal AI. There was an important investigation published by ProPublica in 2016 showing that COMPAS incorrectly flagged Black defendants as likely to re-offend 2 times more frequently than White defendants and incorrectly flagged White defendants as likely to reoffend 2 times more frequently than Black defendants. The creator of COMPAS (Equivant, aka Northpointe) disputed this assessment. A subsequent scholarly discussion about how to measure algorithmic fairness revealed
considerable tension between competing definitions of statistical parity.13 Finally, research in legal language models that exhibit bias mirrors these results as well. A
2022 study published in the Stanford Law Review documented that LLMs grounded on
legal corpora generate biased narratives when writing character references and sentencing memoranda. The bias displayed was that less sympathetic narratives were developed for the named defendants that had a name associated with a minority race. 6. 1. Data Privacy and Cyber security- According to the rules governing attorneys as per Rule 1. 6 of the Model Rules, there is a fundamental duty of confidentiality when it comes to information about clients. Many commercial AI contract review and research tools require attorneys to upload their client documents to cloud-based AI platforms for review, which can impact the attorneys’ confidentiality requirements in complex ways. 14 Attorneys must look to the AI vendor’s data security and methods for handling their clients’ confidential information in order to fulfill their professional obligations under the Model Rules.
Certain types of AI vendors (for example, early publicly available large language models) utilized the data submitted to their models as training data for their models, therefore potentially including confidential information about clients into the AI vendor’s 13 W. Bradley Wendel, Lawyers, Citizens, and the Internal Point of View, 75 Fordham Law Review 1473 (2006) 14 Alexandra Chouldechova, Fair Prediction with Disparate Impact: A Study of Bias in Recidivism Prediction Instruments, 5 Big Data 153 (2017)
database that can be accessed by third parties. Several state bar associations have formally weighed in on the use of AI vendors’ platforms with clients’ data and require that lawyers conduct due diligence on the data handling requirements of the vendors before they engage with them to use their platforms. For example, in its 2024 opinion, the Ethics Committee of the New York State Bar Association stated that lawyers using AI tools have an ethical obligation to take “reasonable measures” to mitigate the risks associated with any unauthorized disclosures of client information and that this includes reviewing the vendor’s terms and conditions of service, as well as negotiating service contracts with the vendors.15
7. Regulatory Frameworks-
The NIST has released an AI Risk Management Framework in 2023. This framework is voluntary and applies to legal contexts for managing AI risks. President Biden has issued an Executive Order on Safe, Secure and Trustworthy AI in October 2023. This executive order directs every federal agency, and each sector, to develop guidance specific to AI. However, these efforts were essentially halted due to an administration change. The European Union has established the most comprehensive regulatory framework regarding AI, specifically through the EU AI Act. 16The EU AI Act went into effect on August 2024. It establishes a risk-based regulatory approach to AI, regulating 'high-risk' AI systems. High-risk AI systems include systems that are used for law enforcement, migration, and the administration of justice.
All high-risk systems must go through a conformity assessment process, maintain technical documentation, have human oversight, and register with the EUR to go into operation. 17 There are several standing orders from federal courts in the U. S that govern the use of AI in litigation. The most significant of those orders is the one from Judge Brantley Starr, Northern District of Texas, requiring an attorney certifying to the Court that any filing using AI has been checked for accuracy by a human being, and that any citations generated by
15 Rashida Richardson(ed.), Confronting Black Boxes: A Shadow Report of the New York City Automated
Decision System Task Force, A.I Now Institute (14 Dec. 2019)
16 Supra note 10
17 Laurence Diver, Digisprudence: The Design of Legitimate Code, 13 Law, Innovation, and Technology 325 (2021)
AI are verified.18
8. Future of Artificial Intelligence in the Legal Field-
There are many complex issues related to the economics of the legal profession. According to a 2023 Goldman Sachs analysis, AI could automate around 44% of legal tasks lawyers currently carry out, which would make law one of the most vulnerable professions in terms of potential job loss due to AI. However, the report also differentiated between the automation of more legal task job loss, indicating that many of the productivity gains from using AI may actually lead to increased output of legal services instead of a decrease in employment levels for litigation professionals. The degree of impact on employment will vary based on a lawyer's seniority and occupation. Those in junior positions with less experience may be at the greatest risk, especially in areas such as routine document reviews, legal research, and due diligence.
19 The traditional path through which junior associates gain practical skills while in training at law firms (essentially via clerical duties and/or working under more experienced attorneys) will be impacted since AI replacing the tasks associated with training first-year attorneys will require law firms to rethink how they will develop associates. However, by utilizing AI, new categories of lawyers are emerging, including legal technology specialists, AI compliance lawyers, and data privacy counsel, as well as algorithmic accountability specialists, who will help create and implement AI engineering systems within the legal framework. Additionally, law schools are introducing dual-degree programs that integrate computer science, data science, and IT management with law. 20
9. Conclusion-
The professional response to artificial intelligence (AI) will be guided by the central principles of the legal profession: competence, client service, justice, and supporting the rule of law. For lawyers, this means not only gaining competence in using artificial 18 Richard Susskind & Daniel Susskind, The Future of the Professions: How Technology Will Transform the Work of Human Experts, 41-68 (Oxford University Press, 2015) 19 Joseph Briggs & Devesh Kodnani, The Potentially Large Effects of Artificial Intelligence on Economic Growth, Goldman Sachs Economic Research (Mar.26, 2023) 20 Ryan Calo, Artificial Intelligence Policy: A Primer and Roadmap, 51 UC Davis Law Review 399 (2017)
intelligence but also acquiring the judgment necessary to evaluate and validate the outputs of AI and to understand the limitations of AI. Bar associations and the courts must develop regulatory frameworks for the rapidly developing technologies that will cover sufficient territory while providing solid protection for the core values of our legal system. Legal educators need to start looking at how they will teach law in a world where machines do most of the routine cognitive activities performed by those in their first few years of practice. The larger question presented by artificial intelligence in law is not about if attorneys will use AI but what does it mean to practice law and do justice in a world of intelligent machines?
The answer to that question will be produced not by the technology but by the decisions made by attorneys, judges, educators, lawmakers, and citizens about how to use, limit, and regulate artificial intelligence relatively to the fundamental values of law.