AI and Pro Se Litigants

I picked up a number of new subscribers from my posts on John Eastman, the leak of documents about the origins of the Supreme Court’s shadow docket, and Ed Martin, and that’s great (welcome to the Substack, y’all!), but I want to reinforce the scope of this newsletter as going beyond Trump administration outrages. For a while I’ve been meaning to write something about the Nippon Life lawsuit against OpenAI.

Most readers of this Substack have probably heard about the case. I’m a bit late to this party and a lot has been written about the lawsuit already. In a nutshell, Graciela Dela Torre was a holder of a long-term disability policy issued by Nippon Life Insurance Company. The company terminated her disability benefits, claiming she was no longer disabled by carpal tunnel syndrome and tennis elbow. Represented by two lawyers, she sued, and eventually the lawsuit settled, with Ms. Dela Torre signing a release of all claims against the insurer. She subsequently became dissatisfied with her lawyer’s performance and told him she wanted to reopen the litigation. The attorney said, nope, you signed a release. A deal’s a deal, you know. In response, Ms. Dela Torre uploaded the lawyer’s response to ChatGPT and “asked whether she was being gaslighted” (Complaint ¶ 50). ChatGPT, being ever so eager to please users,

[a]nalyzed the response and determined that [the lawyer’s] response invalidated Dela Torre’s feelings, dismissed her perspective, and deflected responsibility for her dissatisfaction. ChatGPT ultimately concluded that the tactics used in [the lawyer’s] response constituted gaslighting and were aimed at emotionally manipulating Dela Torre.

Dela Torre then fired her lawyers and started using ChatGPT to prepare a barrage of legal filings, including a motion to vacate the settlement under Fed. R. Civ. P. 60(b). Following the initial motion Dela Torre filed 21 motions, a subpoena, and 8 notices of various sorts (Complaint ¶ 60). She also added Nippon Life as a named defendant to a second lawsuit, and in that action filed 44 motions, demands, and other requests requiring a response from Nippon Life (Complaint ¶ 66). To add insult to injury, ChatGPT included a hallucinated case in one of her filings (Complaint ¶¶ 72-74).

I’ve litigated against pro ses and know many other lawyers who have. Courts generally bend over backwards to give them the benefit of the doubt and ensure that there isn’t a meritorious claim lurking in there someplace. To state the extremely obvious, however, ChatGPT is a major force multiplier for pro se litigants. (See this article about the impact of AI on employment litigation.) Uploading some correspondence and entering the right prompts is enough to produce a blizzard of plausible-looking filings that have to be reviewed with some care by opposing counsel and the court. The complaint includes a lot of what turns out to be AI slop, but look like pretty convincing approximations of real legal filings. As the Complaint says, near the end (¶ 105):

As a lay person with no legal experience . . . Dela Torre would not have been able to . . . file at the same volume and frequency but for the legal assistance provided by OpenAI through its ChatGPT application.

I think that’s very true, but the question here is what did OpenAI do wrong?

The complaint makes a big deal out of Ms. Dela Torre’s motivations. It says the filings “reveal a tone of resentment and personal hostility” (¶ 79), “reveal[] a state of mind driven by sustained animosity rather than any objective legal purpose” (¶ 80), and indicate that she “views this litigation as a vehicle to express and act on her anger toward Nippon” (¶ 81). The Complaint also goes into considerable detail to show that her filings, including those directed at a Nippon in-house lawyer, were legally improper or made for an improper purpose (¶¶ 84-103). Okay, but . . . isn’t this a lawsuit against ChatGPT? What does Ms. Dela Torre’s motivation have to do with the tech company’s liability?

There’s obviously a huge policy issue here, concerning the cost of legal services, the resulting gap in access to justice, and the potential of generative AI tools to assist pro se litigants in prosecuting valid claims. But being a torts as well as PR teacher, I want to take a more nitty-gritty approach and look at the allegations in the lawsuit against OpenAI to see whether they amount to anything, doctrinally speaking, whatever we may think about the potential benefit of making legal services more broadly accessible using GenAI or other technological means.

The question is, therefore, could the maintenance of a consumer-facing, LLM-based, generative AI product that could be used to answer questions about the law, second-guess advice given by a lawyer, and generate legal documents including pleadings and motions, constitute (1) tortious interference with contract, (2) abuse of process or malicious prosecution (more or less the same thing), and (3) unauthorized practice of law, if used by a pro se litigant to hound the bejesus out of an opposing party in litigation?

One reason I’m writing this is in response to a lot of the reporting on the case, which tends to see it primarily as about unauthorized practice of law (UPL). This Bloomberg article, for example, says:

The central question in the Nippon case is whether OpenAI, by suggesting a legal strategy and generating documents to be filed in court, engaged in unauthorized practice of law.

Ehh, I don’t know, or at least I’d refine that a bit. As a torts person, I think the central question may be whether it’s possible to piggyback UPL onto an interesting – and potentially dangerous – common law claim against OpenAI or a similar tech company by linking it with an established cause of action like tortious interference. By itself, UPL doesn’t have much, if any, of a liability punch. Combined with another basis for a tort claim, however, the willy-nilly provision of legal advice and document preparation services by ChatGPT, Claude, and other GenAI tools may expose these companies to liability, or at least the headache of having to defend these lawsuits.

Under Illinois law, a claim for tortious interference with contract requires that the plaintiff [Nippon Life] establish (1) the existence of a contract [here, the settlement of Ms. Dela Torre’s initial claim], (2) defendant’s [Open AI’s] knowledge of the contract; (3) defendant’s intentional and unjustified inducement of a breach; (4) subsequent breach by the defendant; and (5) damages. See Complaint ¶ 108; see also this article.

The knowledge and intent elements both seem difficult to satisfy here. The Complaint addresses knowledge by pleading facts that Ms. Dela Torre fed all of her attorney’s correspondence into ChatGPT, including the attorney’s admonition that she had entered into a contract with the insurer (Complaint ¶ 53 (“ChatGPT was therefore aware of the existence of the agreement.”). But that just means this information was an input into a complex statistical function that outputted the legal documents filed by Ms. Dela Torre. I’m not an epistemologist, and I’m certainly not a computer scientist, but I don’t think information that is used by a computer to calculate a probability distribution and, based on that, output a document, can be equated to the computer “knowing” the information.

As to intent, the Complaint makes a couple of arguments: First, OpenAI has multiple layers of surveillance to monitor ChatGPT usage to prevent harmful conduct (¶ 30). Second, ChatGPT was intentionally designed with features that allow users to obtain legal advice and legal documents from the product (¶ 32). But I’m not sure these satisfy the intent element at the right level of specificity. In general, intent in both criminal and tort law, is something that individual natural persons have, not corporations. The intent of individual human agents may be imputed to a corporation for the purposes of criminal liability, but it kind of doesn’t make sense to say that ChatGPT intended either to practice law or to interfere with the contract between Nippon Life and Ms. Dela Torre. It’s the latter intent that matters for the purposes of tort liability. As I try to get my students to see in first-year Torts, working with the classic case of Garratt v. Dailey, getting too loosey-goosey with the intent element of an intentional tort risks transforming intentional torts into negligence. It could very well be that OpenAI is negligent with respect to the uses to which ChatGPT is being put, but it seems like a stretch to say it (or any human agents of the corporation) had any knowledge at all, let alone intent, with respect to the insurer-policyholder settlement agreement.

The intent element also includes a sub-element that the inducement to breach be unjustified. It’s more common to see that element stated in terms of the use of wrongful means to induce the breach. (That’s how I teach tortious interference on the rare occasion when I teach commercial torts.) A law review article from the 1980s, which post-dates most of the cases cited in the Complaint, states that “interference often is carried out by means of a wrongful act which in itself constitutes an independent tort.” The most recent case cited in the Complaint, Webb v. Frawley, 906 F.3d 569 (7th Cir. 2018), glosses element #4, above, as breach of contract “caused by the defendant’s wrongful conduct.” Wrongful conduct amounting to an independent tort includes things like making defamatory statements or threatening physical violence.

What about engaging in the unauthorized practice of law? If ChatGPT is, in fact, committing UPL, then is the conduct “wrongful” for the purposes of element #4 of tortious interference? Many observers of this case have noted that the Illinois UPL statute does not create an implied right of action. That’s right, but maybe UPL is an instance of wrongful conduct, so it can be used to satisfy that element. The Complaint cites a Ninth Circuit copyright infringement case which also includes a tortious interference claim – MDY Industries v. Blizzard Entertainment, 629 F.3d 928 (9th Cir. 2010).1 That case, in turn, cites the Second Restatement of Torts, Section 767, for seven factors that can be used in determining whether conduct is wrongful: (1) the nature of the defendant’s conduct; (2) the defendant’s motive; (3) the plaintiff’s interests with which the defendant interfered; (4) the interests the defendant sought to advance; (5) the social interests in protecting the defendant’s freedom of action and the plaintiff’s contractual interests; (5) the proximity or remoteness of the defendant’s conduct to the interference; and (7) the relation between the plaintiff and the defendant.

This just confirms my disdain for these multi-factor tests, because really, does that list provide any guidance at all in determining whether interference via unauthorized practice of law is wrongful? With all of these tests the trick is to boil them down into something administrable. This test seems to invite a comparison between the culpability of the defendant’s conduct and the importance of the contract. The finality of settlements is pretty important, which is why they can be undone only on narrow grounds. That cuts in favor of finding OpenAI’s conduct wrongful. On the other side, it’s undoubtedly important to protect access to information about the law, which cuts against a finding of wrongfulness.

My hunch – and isn’t that the problem with multi-factor tests, that they invite decision on the basis of hunches? – is that a court would not regard UPL as sufficiently wrongful to support a tortious interference claim. But I’m not the judge in this case and I could see a judge going the other way. There’s some danger here for OpenAI. If any of the claims in this complaint succeeds, I would expect it to be this one.

The tort of abuse of process in Illinois has two elements:

The two distinct elements of an abuse of process claim are: (1) the existence of an ulterior purpose or motive and (2) some act in the use of process that is not proper in the regular course of proceedings.

Courts say that it is a disfavored claim, so the elements are strictly construed.

The law of many states places severe restrictions on the cause of action for abuse of process (and its close cousin, malicious prosecution). They do so for reasons that are familiar from the debate around Fed. R. Civ. P. 11 as a constraint on frivolous litigation: Courts are understandably worried about what I call the Bleak House problem, where the underlying lawsuit gets buried by ancillary litigation over whether some pleading or contention is not adequately supported. Better to let courts sort it out on the merits, rather than diverting judicial resources into litigation over the propriety of the pleading. Courts are also concerned about upsetting the balance between deterring meritless litigation and providing access to judicial remedies for harms suffered as a result of wrongful conduct. For these reasons, there are frequently significant hurdles to clear for a plaintiff alleging malicious prosecution or abuse of process.

New York law, for example, imposes a strict “special injury” requirement as an element of pleading abuse of process: A plaintiff must show a particularized interference with a person, property or business; this will generally be in the form of a provision remedy such as an “arrest, attachment, injunction, receivership, or notice of pendency.” The special injury must be “some concrete harm that is considerably more cumbersome than the physical, psychological or financial demands of defending a lawsuit.” The headache of defending Ms. Dela Torre’s lawsuit is nowhere close to satisfying this requirement.

As it happens, however, Illinois law takes a looser approach to abuse of process and does not impose anything like the New York special injury requirement.2 In Kumar v. Bornstein, 820 N.E.2d 1167 (Ill. App. 2004) (cited in the Nippon Life complaint), an Illinois appellate court surveyed the law on abuse of process, found that all of the successful cases involved something like a prejudgment arrest or attachment, but denied that this was a necessary condition on recovery:

While we have found no Illinois case, and plaintiff cites none, that allows an abuse of process claim in the absence of an arrest or physical seizure of property, we agree with plaintiff that an arrest is not a required element of abuse of process. Abuse of process is defined as the misuse of legal process to accomplish some purpose outside the scope of the process itself. Bonney v. King, 201 Ill. 47, 50-51 (1903). The only elements necessary to plead a cause of action for abuse of process are: (1) the existence of an ulterior purpose or motive and (2) some act in the use of legal process not proper in the regular prosecution of the proceedings. Holiday Magic, Inc. v. Scott, 4 Ill. App. 3d 962, 966 (1972).

Okay, but I’m not sure that’s going to help Nippon Life here. When I teach abuse of process and malicious prosecution (more likely to come up in PR than in Torts), the big theme is that there has been a systemic shift away from these common law doctrines toward sanctions administered by the court before whom the litigation is pending, based on Fed. R. Civ. P. 11, 28 U.S.C. § 1927, and the court’s inherent power. Those doctrines perform similar functions and, to some extent, mirror the torts of abuse of process and malicious prosecution. Importantly, however, there is a significant transaction cost advantage in relying on courts supervising litigation, rather than a separate, after-the-fact lawsuit, to determine whether a claim has merit or whether it was brought for an improper purpose. That’s why you see courts referring to abuse of process and malicious prosecution as disfavored causes of action. (See these excellent CLE materials on these claims by Hinshaw & Culbertson.)

That’s not to say there aren’t cognizable abuse of process claims out there, but the language quoted above is important: Tort claims generally require that the legal system be used for some purpose outside the scope of the process itself. Obsessive and annoying as she was, Ms. Dela Torre was basically using the legal system to pursue a legal claim. She may have been egged on by ChatGPT into thinking that she had a valid claim when she did not, but she wasn’t doing something different in kind from doggedly pursuing a claim.

This doesn’t mean her conduct wasn’t wrongful, but it was wrongful-qua-litigation-sanctions, meaning that from the systemic point of view, the best response is to leave it to the courts supervising these actions to decide to impose sanctions, or not. Rather than deal with non-meritorious or even abusive litigation with an after-the-fact tort action, better to leave the control of this misconduct to the court before whom it occurred. Courts don’t tend to sanction pro se litigants, for the understandable reason that unrepresented people don’t have the same level of understanding of the legal and factual support necessary to make a claim meritorious.

And that’s before we get to the real problem with this claim, which is that, like tortious interference with contract, it is an intentional tort. There’s a nice example of legal spin in ¶ 20 of the Complaint, where is says “Courts in the United States have long held that the liability for an abuse of process claim extends to all who knowingly participated, aided, or abetted the abuse, as well as those who advised or consented to the unlawful act.” That sounds like support for casting a wide liability net – wide enough to include OpenAI. But go back and read that passage and notice the word “knowingly.” Again, providing a software tool that can be used for good or ill, which is then used for ill by a consumer, is not the same thing as knowingly providing assistance in the unlawful act.

As far as I can tell – and I’m not an expert on Illinois tort law – Illinois is somewhere between California, which is fairly friendly to abuse of process claims, and New York, which is decidedly hostile. Still, it’s a disfavored claim, even in Illinois. I don’t think this is a frivolous claim for abuse of process, but I think it’s ultimately a loser given that it’s a disfavored cause of action, that the courts dealing with Ms. Dela Torre’s blizzard of motions had a chance to sanction her and chose not to, and the difficulty proving the intent element.

Illinois, like all other states, defines the practice of law as including much more than appearing on behalf of another in a litigated matter. A summary prepared by the Illinois Bar Association states that the practice of law includes “giving advice or services ‘when the giving of such advice or rendition of such service requires the use of any degree of legal knowledge or skill.’” (Citing People ex rel, Illinois State Bar Association v. Schafer, 404 Ill. 45, 50 (1949)) The provision of general information about the law is not the practice of law, but if the information, advice, or product (such as a letter or pleading to be filed in court) is responsive to the particular situation of another person, it could very well be the practice of law. As a recent blog post on the Georgetown Journal of Legal Ethics site observes:

AI’s functional mechanisms, driven by algorithmic responses to unique user inputs, muddle the question of what constitutes mere resource provision versus informed counsel. Generative AI systems produce responses based on statistical patterns in training data rather than professional judgment. Yet when an AI system answers a user’s legal question with specific guidance about a particular case, the difference between general information and personalized advice begins to blur.

I would go farther and say the line isn’t just blurred, but it has been crossed when the AI system generates output based on user inputs with specific questions about legal rights and procedure, particularly when accompanied by input of legal reasoning generated by a lawyer.

Just because something is UPL, however, doesn’t mean there is a private cause of action for UPL. The Illinois Attorney Registration and Disciplinary Commission (ARDC) has a helpful overview of the enforcement mechanisms available in response to UPL:

[T]he ARDC is not the only player when it comes to UPL enforcement actions. The Attorney General’s office has wide-ranging authority to pursue claims against UPL offenders under a number of different statutes. State’s Attorney’s offices also have the ability to formulate criminal charges for theft or False Personation of an Attorney, 720 ILCS 5/17-2, among other things. Finally, the Attorney Act provides standing for attorneys and bar associations to bring claims against those that practice law without authorization, although such actions require expenditure of considerable time and resources.

What’s not on that list, however, is tort actions by individuals or corporations claiming they were damaged as a result of UPL. The Complaint (¶ 23) doesn’t really dispute this, but instead alludes to the independent power of federal courts to manage their own affairs to achieve the efficient and expeditious resolution of cases. That’s certainly a thing, but it’s duplicative of the abuse of process and litigation sanctions doctrines discussed above, and the company isn’t going to be able to bootstrap a private cause of action off the court’s inherent authority.

My colleague James Grimmelmann, who has been thinking a lot lately about various tort theories of liability against AI systems, pointed me to a case in which an AI was sued for the unauthorized practice of medicine. In that case, however, the lawsuit was brought by the Pennsylvania State Board of Medicine. It’s certainly possible that something similar could happen in Illinois, with the ARDC going after OpenAI. For now, however, I don’t see how Nippon Life is the proper party to allege UPL, although as mentioned above, I can see linking the UPL allegation to a tortious interference claim.

The complaint was filed by Sidley Austin, a serious firm, although they subsequently moved to withdraw as counsel. Although I think the claims will not ultimately be successful, the complaint is well thought out and raises serious issues. (Yes, I thought about the irony that a complaint for abuse of process could itself be abusive.) So what is Nippon Life’s game here? Is this a shot across the bow of AI companies, warning them to take more care to ensure that their products aren’t used for unlawful purposes, like providing legal assistance without a lawyer? Is it meant to highlight a regulatory gap, with AI systems being outside the reach of the lawyer disciplinary process (except for UPL)? I’m waiting to see the response from OpenAI, presumably in the form of a motion to dismiss, and what the court does at the pleading stage. This will be super interesting if it gets past a motion to dismiss and gets into discovery. For now I’m just grateful to have an AI topic to write about that isn’t yet another order to show cause for including hallucinated citations in a brief.