I. Introduction

In his 1998 debut techno-thriller novel, Digital Fortress, Dan Brown poses a compelling question: "Quis custodiet ipsos custodes?"—Latin for Who guards the guards? Brown’s injunction urges the readers to be cautious of the seemingly impenetrable digital software that silently surveils data without consent. Today, what was once confined to fiction has entered personal injury law firms in the form of sensitive information from vulnerable clients' testimony entrusted to black-box algorithms. The crucial issue to consider here is that the duty that we have assigned to AI is far more substantial than the protection of foreign security that Brown referred to in his novel. In fact, it is the protection of human dignity and credibility at a very fragile moment of their lives.

This paper argues that artificial intelligence has not upended legal practice so much as it has reconfigured the economic architecture of civil litigation and redefined the stakes of confidentiality, factual integrity, and judicial credibility. AI can democratize personal injury law by equipping smaller firms with technological adaptability that rivals institutional scale without requiring institutional resources. Nonetheless, whether that democratizing potential can be fully realized depends on firms’ commitment to informed professional oversight and independent legal judgment, as well as their ability to determine which legal tasks must remain nondelegable to a chatbot. Otherwise, AI may reproduce the very inequalities it appears capable of disrupting.

II. The Equalizer

The cause of structural bifurcation in the American civil litigation market is rooted in something much more complex than trial verdicts. It is fundamentally driven by the economics of data processing. In his seminal 1974 framework, Why the 'Haves' Come Out Ahead, legal scholar Marc Galanter attributes the structural divide to the “Repeat Player” advantage. Such institutional “repeat players”, namely large-scale insurance companies and multi-national corporations, have the organizational capacity to absorb litigation costs, strategically delay discovery, and shape negotiations on their own terms.

According to the article “Defining the Problem of Cost in Federal Civil Litigation” from the Duke Law Journal, empirical studies demonstrate that discovery costs can reach up to as much as 70 percent of total litigation costs in some cases. The resource gap between One-Shot Players and Am Law 100 firms manifests in the pre-trial processes in which elite firms use legions of contract attorneys to process thousands of pages of data while the small boutique firms use their entire legal staff to review only a fraction of that volume. Thus, for decades, the corporate retainers or “Repeat Players” weaponize procedural bureaucracy to secure asymmetric settlements with the rather vulnerable firms, a system of justice in which economic capability undermines the merits of a claim itself.

However, AI has changed the power dynamic through disproportionately benefiting small-scale contingency-based law firms. The ABA Journal has reported that personal injury attorneys are highly incentivized to adopt AI as they use a contingency-fee model and consequently, the efficiency gains immediately improve the attorney’s baseline profit. According to research conducted by the Harvard Law School Center on the Legal Profession, specialized AI tools can shrink certain litigation tasks from 16 hours down to 3–4 minutes. The “BigLaw” firms are now burdened by bureaucracy in the way that their countless compliance regulations and approval committees delay the large-scale adoption of AI across their firms whereas small personal injury firms are able to deploy the AI tools within weeks. Personal injury lawyer and co-founder of the Simon Law Group exemplifies this reorientation of power in the civil litigation sphere: "You have this group of lawyers who are willing to adopt AI, and they're not tied down by bureaucracy.”

Investment trends reflect this shift. For instance, Supio, a personal injury AI platform that automates the analysis of unstructured medical records, has received a $60 million Series B round in 2025. Moreover, Eve Legal, an AI software that targets historically underfunded firms, secured a $47 million Series A round led by Andreessen Horowitz. Through my law internship at a local firm, I have seen the drastic effects of platforms like Eve have in real-time: the firm I worked at, one with just two in-house attorneys, was able to merge with a 300-person law firm partially because of their technological agility, a trait that has become its own type of legal capital.

AI has created an operational paradigm in the litigation market in which the right AI infrastructure rivals institutional scale, enabling small firms to not only catch up with larger practices but to become an asset that larger firms fight over. Contrary to software licenses which any firm can obtain, small practitioners use their technological adaptability to strategically train staff and maximize their contingency-based client pipeline. Since building these systems internally is far more capital-intensive than acquiring such a firm, the “Repeat Players” no longer see these small practices as competitors, rather they approach them as opportunities for acquisition. More importantly, such legal tech has shifted the competitive axis of civil litigation away from depth of the bench to the legal value of the case.

However, the democratization of civil litigation power has given way to legal vulnerabilities as well. Although generative AI may allow boutique plaintiff-side firms to compress workflows, which once required weeks of paralegal labor, that rate of productivity can also be used against them. This newly adopted efficiency clashes with the moral boundaries of ABA Formal Opinion 512 which requires that fees must remain “reasonable” (standard rule Model 1.5) regardless of whether AI can complete research that historically required weeks in only minutes. Ethics expert Mark C. Palmer from the Illinois State Bar Association’s Artificial Intelligence Committee warned that computational efficiency would threaten the dominant billable hour model, coining this ethical complication as the “AI efficiency paradox." The dominant contingency fee that most personal injury firms use is not directly affected; nonetheless, corporate defendants manage to leverage opposing counsel’s use of AI as a Trojan horse to lower attorney fees.

This blueprint is exemplified by Uber’s multi-state campaign for policy change in personal injury law. In response to several sexual harassment and personal injury lawsuits, Uber has proposed ballot initiatives that push for sweeping tort reform such as California’s “Protecting Automobile Accident Victims from Attorney Self-Dealing Act” and the Nevada Limit Contingency Fees in Civil Cases Initiative, all of which make it economically difficult for small boutique firms to pursue their practice. Uber’s PR apparatus frames the campaign as a consumer-protection measure that allows plaintiffs to profit significantly more from their recovery. Even though Uber does not explicitly weaponize AI as a way to lower attorney fees, its message that attorneys are overcompensated becomes increasingly persuasive in an AI era. Although these initiatives are not AI-driven, they capitalize on a growing public perception that technology has reduced the value of legal labor. The corporation has the ability to exploit the public’s misconception that AI prices out elite advocacy when in reality, plaintiff firms must pay substantial upfront litigation costs for matters including but not limited to discovery and hiring experts that AI can’t offset. Thus, the widespread skepticism on the value of legal labor is shaping the culture of attorney compensation to Uber's benefit regardless of whether the company codifies AI into their campaign. Simultaneously, Uber can use AI against these very plaintiffs as a way to discredit their claims.

This corporate hypocrisy is crystallized in the federal judiciary. In MDL No. 3084 in the Northern District of California–a massive multidistrict litigation regarding an Uber passenger sexual assault claim–the corporate defense counsel filed a sanctions motion, alleging that the plaintiffs potentially used inaccurate AI-generated citations in their discovery in the bellwether case. Uber’s capped-fee campaigns benefit from the AI-era suspicion on overcompensation yet they simultaneously attack AI’s factual inaccuracies to ruin the credibility of the plaintiff–revealing a dual logic in their defense apparatus. Thus, artificial intelligence which seems to be the equalizer in the litigation market has the capability to absorb into the very asymmetries it aimed to upend. 

III. Confidentiality Crisis

Although generative artificial intelligence has led to the redistribution of legal power, it has also brought a renewed focus to a doctrine at the heart of common-law tradition: evidentiary privilege. In other words, the privilege that certain confidential communications are protected from being forced into evidence. The application of AI carries risks that are far more serious than data leakage complications. In fact, AI has the capacity to jeopardize the confidentiality rules at the heart of Anglo-American jurisprudence through its black-box architecture and large language model mechanics.

Under ABA Model Rule 1.6, an attorney has the duty to protect information related to the representation of the client from unauthorized disclosure. Yet, this rule becomes considerably less clear when taking into account the question of whether communications with AI tools remain protected from discovery. In United States v. Heppner, Judge Jed S. Rakoff of the Southern District of New York ruled that a defendant’s interactions with the consumer version of Claude may strip the client the client of any “reasonable expectation of confidentiality” given that Anthropic specifically says it may disclose user information in response to legally valid government/law enforcement requests for user information. Thus, the client’s self-directed prompts to Claude could be sought for discovery as Judge Rakoff reasoned that “all recognized privileges require a trusting human relationship” with a licensed professional who owes fiduciary duties and Claude does not fit the criterion to form a "privileged relationship.” As exemplified by this very ruling, the use of AI is both a legal tool and legal vulnerability. Its risks go far beyond the data leakage warned against in ABA Formal Opinion 512. In fact, it risks waiving privilege itself.

The parallel with Brown’s Digital Fortress is more than literary flourish as it raises the question at the heart of the confidentiality crisis: who guards the systems that are entrusted to a client’s personal data? An attorney who uploads a plaintiff’s unstructured medical information and personal data in exchange for computational speed makes a dangerous Faustian bargain in which the privilege risks are not necessarily protected. Conversely, attorneys who answer Brown’s question in their very work through adopting an AI-based ring-fenced infrastructure minimize privilege risks through safeguarding all data from commercial networks. In fact, I witnessed this shift in real time at my local internship: the firm deployed Eve Legal which functions as a “closed-loop” enterprise. In their security and compliance statement, Eve guarantees “zero-retention API” where no prompts/documents are used to train AI models. As shown, using AI ethically and legally requires two components: a "digital fortress” that protects all sensitive data and an attorney who is well-aware of AI’s capacity to fortify or fracture a case.

IV. Hallucinations

Aside from confidentiality concerns, the algorithmic integration introduces a new risk: the tendency of large language models to provide authoritative-sounding precedent and evidence that fails to exist. A Stanford University RegLab study published by the Journal of Empirical Legal Studies exemplifies these data flaws, finding hallucination rates of 33 percent for Westlaw AI-Assisted Research and 17 percent for Lexis+AI. Moreover, in the landmark companion study, “Large Legal Fictions”, researchers confirmed the hallucinations occur between “58% of the time with ChatGPT 4 and 88% with Llama 2.” Thus, the adoption of AI introduces a new asymmetry. While BigLaw firms have the capacity to deploy contract attorneys to audit and manually verify AI responses, small boutique firms may replace all paralegal labor with AI which exposes new ethical liabilities that can jeopardize their credibility.

The Stanford researchers also addressed the Retrieval Augmented Generation (RAG) model, a system that connects a large language model with a specific case law database as a way to prevent all fake citations. Notwithstanding, the AI is oftentimes overwhelmed by the information and uses the incorrect sources to draft responses. The solution of fabricated outputs is not suppressing the innovation entirely, but using systems that narrow their retrieval focus. For instance, Eve Legal contains all of its research within the firm’s data ecosystem and localizes all verification rather than using the standard RAG model, ultimately minimizing any fake citations and hallucinations. Thus, the application of legal AI in personal injury firms are only as useful as their limits: the boundaries of a chatbot’s search determine whether an attorney maintains factual integrity or fictionalizes precedent before a judge.

AI-generated hallucinations have also led to the erosion of the court’s truth-seeking function, an institutional dilemma crystallized by the Mata v. Avianca Inc. (2023) case wherein the judge imposed Rule 11 sanctions after both attorneys submitted AI-generated fabricated federal cases. At first, expert commentary dismissed the case as an anomaly of irresponsibility. In contrast, legal data scientist Damien Charlotin has documented more than 1,200 cases of hallucination sanctions globally as of 2026–exemplifying how the structural dangers of AI have gone from isolated reprimands to a rather pervasive contagion that challenges judicial integrity.

The crisis reached an inflection point in the case of Withers v. City of Aberdeen (2026) wherein Judge Sharion Aycock underscored that an attorney can not substitute human judgment with computational verification through sanctioning both the defendant and plaintiff counsels after they submitted briefs with nonexistent case law. Aycock sent a clear message: the ethical responsibility to verify information cannot be delegable to a chatbot. When a small boutique practitioner chooses computational speed over legal analysis, they not only misuse an instrument meant for efficiency but they corrode both their ethical standing and the judiciary’s legitimacy.

This hallucination problem transcends the misuse of case citations, it lies in the distortion of legal analysis and synthesis. According to legal scholars writing for the Canadian legal publication Slaw in 2026, the deeper threat is not the obvious fabrication of case law. It is “analytical misalignment.” Even if artificial intelligence platforms can minimize inaccuracies through RAG models and close-loop infrastructure, the flaw lies in the fact that the chatbot lacks competent legal judgment and oftentimes misinterprets the case law. Through applying the logic and rationale of AI, lawyers sacrifice their ethical duty of competence under Model Rule 1.1. The silent attrition of analytical precision extends far beyond the visible sanctions docket, it can undermine a firm’s credibility from within.

V. Conclusion

Artificial intelligence has the capacity to reduce legal stratification by allowing small boutique firms to challenge institutional ‘Repeat Players’; however, when such computational speed begins to erode factual credibility and the truth-seeking function of the judiciary, AI can be absorbed into the very corporate inequalities it promised to flatten. Dan Brown's message in Digital Fortress wasn't criticizing the instability of powerful systems, rather it was exposing how vulnerable humans are when they become overly reliant on certain systems that are seemingly "impenetrable." This same peril has now woven itself into the soul of the legal profession. A digital fortress without legal acumen and human advocacy is not a secure hub for mechanized productivity. It is simply another black box.

VI. Acknowledgement

The Institute for Youth in Policy wishes to acknowledge Michelle Young for editing this policy brief.

VII. References

“- YouTube.” Www.Youtube.Com, https://www.youtube.com/watch?v=vbjlpxc2c7k. Accessed 21 July 2026.

American Bar Association. “Rule 1.6: Confidentiality of Information.” Americanbar.Org, 2019, https://www.americanbar.org/groups/professional_responsibility/publications/model_rules_of_professional_conduct/rule_1_6_confidentiality_of_information/. Accessed 21 July 2026.

American BAR Association Standing Committee on Ethics and Professional Responsibility. 2024, https://www.americanbar.org/content/dam/aba/administrative/professional_responsibility/ethics-opinions/aba-formal-opinion-512.pdf. Accessed 21 July 2026.

Brown , Dan. Digital Fortress.

Charlotin, Damien. “AI Hallucination Cases Database.” Damiencharlotin.Com, 2025, https://www.damiencharlotin.com/hallucinations/. Accessed 21 July 2026.

Couture, Robert. “The Impact of Artificial Intelligence on Law Firms’ Business Models - Harvard Law School Center on the Legal Profession.” Harvard Law School Center on the Legal Profession, 25 Feb. 2025, https://clp.law.harvard.edu/knowledge-hub/insights/the-impact-of-artificial-intelligence-on-law-law-firms-business-models/. Accessed 21 July 2026.

Fonarev, Jamie. “What Does It Mean to ‘Trust’ Your AI?” Eve.Legal, Eve, 26 July 2024, https://www.eve.legal/blogs/what-does-it-mean-to-trust-your-ai. Accessed 21 July 2026.

Freedman, Robert. “Uber Turns Nevada into Contingency Fee-Cap Battleground.” Legal Dive, 25 Sept. 2024, https://www.legaldive.com/news/uber-turns-nevada-into-contingency-fee-cap-battleground-gupta-tort-reform/728094/. Accessed 21 July 2026.

Guest Blogger. “Trust No AI? : Updating the Duty of Competence for the Modern Lawyer.” Slaw, Feb. 2021, https://www.slaw.ca/2021/02/01/trust-no-ai-updating-the-duty-of-competence-for-the-modern-lawyer/. Accessed 21 July 2026.

“In Re: Uber Technologies, Inc., Passenger Sexual Assault Litigation.” Northern District of California | United States District Court, 4 Oct. 2023, https://cand.uscourts.gov/cases-e-filing/cases/323-md-03084-crb/re-uber-technologies-inc-passenger-sexual-assault-litigation. Accessed 21 July 2026.

“ISBA Ethics Opinions on Contingent Fees | Illinois State Bar Association.” Isba.Org, 2026, https://www.isba.org/ethics/subjects/contingentfees. Accessed 21 July 2026.

Kumar, Mira. “United States V. Heppner.” Harvard Law Review, 23 Mar. 2026, https://harvardlawreview.org/blog/2026/03/united-states-v-heppner/. Accessed 21 July 2026.

LexisNexis. “REV & LexisNexis Invest in EvenUp, the Leader in AI for Personal Injury Law.” Community, LexisNexis, 7 Oct. 2025, https://www.lexisnexis.com/community/pressroom/b/news/posts/rev-and-lexisnexis-invest-in-evenup-the-leader-in-ai-for-personal-injury-law. Accessed 21 July 2026.

Magesh, Varun, et al. “AI on Trial: Legal Models Hallucinate in 1 Out of 6 (or More) Benchmarking Queries.” Hai.Stanford.Edu, 23 May 2024, https://hai.stanford.edu/news/ai-trial-legal-models-hallucinate-1-out-6-or-more-benchmarking-queries. Accessed 21 July 2026.

“Penalties Stack Up as AI Spreads Through the Legal System.” NPR, 3 Apr. 2026, https://www.npr.org/2026/04/03/nx-s1-5761454/penalties-stack-up-ai-spreads-through-legal-system. Accessed 21 July 2026.

Stanford Law School. “Ai, Liability, and Hallucinations in a Changing Tech and Law Environment | Stanford Law School.” Stanford Law School, 2024, https://law.stanford.edu/stanford-legal/ai-liability-and-hallucinations-in-a-changing-tech-and-law-environment/. Accessed 21 July 2026.

“Are Contingency Fees Competitive? New Paper Says No, but Caps Are No Cure | Stanford Law School.” Stanford Law School, 8 May 2026, https://law.stanford.edu/2026/05/08/are-contingency-fees-competitive-new-paper-says-no/. Accessed 21 July 2026.

Stempel, Jonathan. “US Judge Dismisses Some Claims in Uber Sexual Assault Lawsuits.” Reuters, 9 July 2025, https://www.reuters.com/legal/litigation/us-judge-dismisses-some-claims-uber-sexual-assault-lawsuits-2025-07-09/. Accessed 21 July 2026.

Sumagaysay, Levi. “Uber Ballot Initiative Sparks Showdown with Lawyers, Doctors.” CalMatters, 24 Feb. 2026, https://calmatters.org/economy/2026/02/uber-california-ballot-initiatives/. Accessed 21 July 2026.

Thomas, David. “Nevada Supreme Court Rejects Uber Bid to Cap Lawyer Fees.” Reuters, 27 Jan. 2025, https://www.reuters.com/legal/government/nevada-supreme-court-rejects-uber-bid-cap-lawyer-fees-2025-01-27/. Accessed 21 July 2026.

Vigdor, Neil. “Judge Punishes 4 Lawyers After Catching Both Sides Using A.I. In Lawsuit.” The New York Times, 9 June 2026, https://www.nytimes.com/2026/06/09/us/ai-lawyers-sanctioned-mississippi.html. Accessed 21 July 2026.

Brown, Dan. Digital Fortress. St. Martin’s Press, 1998.

American Bar Association Section of Litigation. Member Survey on Civil Practice: Detailed Report. 11 Dec. 2009. United States Courts, https://www.uscourts.gov/file/document/aba-section-litigation-survey-civil-practice. Accessed 21 July 2026. 

Galanter, Marc. “Why the ‘Haves’ Come Out Ahead: Speculations on the Limits of Legal Change.” Law & Society Review, vol. 9, no. 1, 1974, pp. 95–160. Cambridge Core, https://doi.org/10.2307/3053023.

Hacker, Avi. “AI Hallucinations Trigger Record Court Sanctions: What Legal AI Risk Means for CRE Investors.” The AI Consulting Network, 17 Apr. 2026, https://www.theaiconsultingnetwork.com/blog/ai-hallucinations-legal-sanctions-record-cre-investors-2026. Accessed 21 July 2026.

Lee, Emery G., III, and Thomas E. Willging. “Defining the Problem of Cost in Federal Civil Litigation.” Duke Law Journal, vol. 60, no. 3, 2010, pp. 765–88. Duke Law Scholarship Repository, https://scholarship.law.duke.edu/cgi/viewcontent.cgi?article=1486&context=dlj. Accessed 21 July 2026. 

Mata v. Avianca, Inc. 678 F. Supp. 3d 443. U.S. District Court for the Southern District of New York. 22 June 2023. Berkeley Law, https://www.law.berkeley.edu/wp-content/uploads/archive/2025/12/Mata-v-Avianca-Inc.pdf. Accessed 21 July 2026.

Robert, Amanda. “Personal Injury Lawyers Find ‘Massive Opportunity’ with AI.” ABA Journal, 4 Dec. 2025, https://www.abajournal.com/web/article/personal-injury-lawyers-find-massive-opportunity-with-ai. Accessed 21 July 2026.

Supio Team. “Supio Announces $60M Series B to Accelerate Adoption of Legal AI in Plaintiff Law.” Supio, 30 Apr. 2025, https://www.supio.com/press/supio-announces-60m-series-b-to-accelerate-adoption-of-legal-ai-in-plaintiff-law. Accessed 21 July 2026.

“Court Weighs Uber’s Sanctions Request in Sexual Assault Bellwether Case.” Daily Journal, 1 June 2026, https://www.dailyjournal.com/article/391787-court-weighs-uber-s-sanctions-request-in-sexual-assault-bellwether-case. Accessed 21 July 2026.

Mazanec, Karel. “Capping E-Discovery Costs: A Hybrid Solution to E-Discovery Abuse.” William & Mary Law Review, vol. 56, no. 2, 2014, pp. 631–72, https://scholarship.law.wm.edu/wmlr/vol56/iss2/7.

Omilaev, Igor. Robot and Human Hands Reaching Toward AI Text. Published September 16, 2023. Unsplash. Accessed July 22, 2026. https://unsplash.com/photos/robot-and-human-hands-reaching-toward-ai-text-FHgWFzDDAOs.