How to Win and Still Lose: Delaware Court of Chancery’s Expanding Focus From Individual AI Misuse to Organizational Governance

A July 1, 2026 order from the Delaware Court of Chancery provides another insight into the risks of using generative artificial intelligence (“GenAI”) without adequate human review, and highlights the court’s increased focus on organizational controls governing AI use. Although Leiske v. Kidd concerned litigation conduct and the duties of attorneys and law firms rather than directors’ fiduciary duties or corporate AI governance, the order suggests a growing judicial focus on not only individual misuse of AI, but also on the processes, controls, and oversight surrounding how AI is deployed within an organization. That focus reflects themes discussed in our May 2026 post, which examined SDNY’s inquiry into organizational questions arising from the use of ChatGPT in government decision-making.

In Leiske v. Kidd, Vice Chancellor Lori W. Will ordered plaintiffs’ counsel and his firm to show cause why sanctions should not issue. The brief at issue contained fictitious citations, fabricated quotations, and inaccurate legal propositions following the use of GenAI to revise the brief. The order emphasizes that an attorney’s obligations under Court of Chancery Rule 11 are nondelegable. But the Court did not stop with the filing attorney. It separately directed the law firm to explain its GenAI policies, training, technological safeguards, supervision, and verification procedures. Vice Chancellor Will’s decision on this matter remains pending, but the case is notable because it illustrates Delaware courts’ willingness to look beyond the erroneous output itself and examine the circumstances of AI use and the controls surrounding it. Leiske does not establish AI-specific fiduciary duties for directors or officers, and Delaware courts have not yet articulated a distinct Caremark framework for AI governance. But as Delaware courts become increasingly familiar with the ways in which organizations integrate AI into their workflows, the questions raised in Leiske may become relevant to cases examining board and officer oversight, internal controls, and organizational responses to emerging risks.

How GenAI Errors Became a Rule 11 Issue

Leiske is an advancement case that was pending before a Magistrate in Chancery and reassigned to Vice Chancellor Will for the limited purpose of resolving exceptions to the Magistrate’s final report. On January 22, 2026, plaintiffs filed an answering brief opposing defendants’ exceptions. Five days later, defendants identified what the Court described as “fictitious citations, fabricated quotations, and hallucinated legal propositions” in the brief. Although plaintiffs ultimately prevailed on the merits of the dispute, the win was overshadowed by plaintiffs’ faulty reliance on GenAI.

Plaintiffs’ counsel acknowledged that a GenAI tool had been used to revise the brief and that the resulting citations had not been verified before filing, attributing the oversight to a paralegal’s review. Plaintiffs filed a corrected brief, but the Court noted that it removed quotation marks from erroneous statements without correcting certain inaccurate descriptions of the underlying law.

Counsel’s response compounded the Court’s concerns. Counsel maintained that the cited authorities supported the legal propositions in the brief and criticized defendants for raising the errors with the Court rather than meeting and conferring. Vice Chancellor Will rejected that position, explaining that there was “nothing to negotiate or compromise” when false citations had been submitted to a tribunal. She concluded that the submission of hallucinated legal propositions raised an inference of a Rule 11 violation and ordered the filing attorney and his firm to show cause why sanctions should not issue.

In response to the order to show cause, plaintiffs’ lead counsel and law firm described the human-review processes and institutional controls that were in place when the brief was filed. Counsel explained that he used an AI-powered search engine to assist in drafting, supplied the tool with selected source materials, and used instructions intended to reduce inaccurate legal output. A paralegal and senior associate also cite-checked the brief, though the lead counsel acknowledged not personally reviewing the citations and quotations before filing. The firm, in turn, described a firm-approved GenAI policy, required training, and review procedures for AI-assisted work, but concluded that these safeguards failed in implementation in this case.

The Court’s decision on the order to show cause remains pending. In the interim, defendants sought permission to make a further submission, asserting that plaintiffs’ counsel’s time records appear inconsistent with the affidavits and offering to present additional evidence concerning the costs defendants incurred as a result of the AI-related errors.

From An to Leiske: Delaware Courts’ Broadening Focus on GenAI Controls

Leiske is not Vice Chancellor Will’s first encounter with GenAI-generated legal errors.

In April 2025, in An v. Archblock, Inc., she addressed a pro se filing containing apparently AI-generated false and miscited legal authority. There, Vice Chancellor Will imposed case-specific requirements governing future GenAI-assisted filings, including disclosure of the tool used and certification that the resulting text and authorities had undergone human review. Because Leiske, unlike An, involved a filing by an attorney, Will’s order invoked Delaware’s Rule 11 requirement that a filing attorney certify that the legal contentions made in the filing are warranted by existing law after reasonable inquiry, and that this duty cannot be satisfied by relying on a paralegal, “much less on an artificial intelligence program.”

Vice Chancellor Will’s order in Leiske went beyond the conditions imposed in An. Considering that Rule 11 makes a law firm jointly responsible for violations committed by its attorneys, she observed that the incident could implicate the firm’s training, supervision, and deployment of GenAI. The firm was therefore required to provide the Court with its written GenAI policies and restrictions, how those policies were communicated, and the internal procedures, technological safeguards, and training programs implemented or planned to prevent similar problems.

The progression from An to Leiske demonstrates the Delaware courts’ continuing inquiry into the organizational use of AI. Whereas An focused on individual misuse and required the attorney to disclose AI usage and human review, Leiske inquired into the employing firm’s organizational framework at the threat of joint liability. Although the case concerns litigation conduct, this organization-focused line of inquiry has potentially broader implications for cases examining corporate governance and oversight of emerging technology risks.

Leiske’s Parallels to American Council

The Court of Chancery’s focus on organizational governance of AI in Leiske parallels SDNY’s approach in American Council of Learned Societies v. National Endowment for the Humanities, as analyzed in our May 2026 post. In that case, government personnel used an LLM to identify federal grants for possible termination based on their relationship to DEI. The Court rejected the government’s effort to distance itself from problematic AI-generated classifications, examined how the tool was integrated into the government’s decision-making process, and focused on deficiencies in prompt design, understanding of the model, human review, validation, and communication.

Several governance lessons identified in our May analysis reappear in Leiske:

• The process for introducing AI into a workflow matters. In American Council, the Court scrutinized how prompts were developed and whether personnel understood how the model interpreted the concepts they asked it to apply. Vice Chancellor Will made a similar inquiry in Leiske: the filing attorney was required to explain who entered the prompts, when and how GenAI was used, and how the output was incorporated into the brief. Both Courts looked upstream from the faulty output to the process that produced it.
• Human review should be meaningful. American Council emphasized the absence of meaningful review before AI-generated outputs became embedded in consequential decisions, finding it significant that no human reviewer had disagreed with an AI-generated rationale and chosen a different result. Leiske presents a parallel issue: although a paralegal had a cite-checking role, the GenAI-generated citations were not verified before filing. The Leiske order required a detailed explanation of the review process, including the instructions given, the tools used for verification, and whether attorneys independently checked the cited text. The focus is not simply whether a review procedure existed, but whether it actually worked (and what the appropriate level of review should be). Indeed, the affidavits filed in response to the order indicate that multiple layers of human review and AI controls failed to prevent the errors from reaching the Court.
• Responsibility remains with the humans and organizations deploying technology. American Council rejected the premise that the government could distance itself from an AI-assisted result by blaming ChatGPT. Leiske reaches a similar conclusion through Rule 11: the filing attorney could not shift responsibility to either a paralegal or GenAI, and by requiring the firm itself to explain its policies, training, and safeguards, the order extends the inquiry beyond the individual to the organization that deployed the technology. The pending sanctions decision may provide further guidance on how courts assess organizational accountability where formal controls existed but failed to prevent the conduct at issue.

Practical Takeaways

AI governance should address systems, not just individual users. The Leiske Court’s focus on policies, training, technological safeguards, supervision, verification, and remediation likely has relevance beyond law firms. Companies deploying AI in consequential business functions should consider whether they can explain not only what their policies require, but also how those policies are communicated, implemented, enforced, and—as relevant to the parties’ affidavits—whether they actually operate in practice.

Delaware courts expect meaningful human review over AI-generated outputs. Organizations should consider who performs that review, what expertise is required, what sources or systems should be used for verification, whether reviewers have authority to reject the output, and how the review process is documented. Both American Council and Leiske illustrate the importance of testing whether human oversight operates as an actual control rather than a nominal step.

Establish clear accountability from AI input through final output. Organizations should understand who may use particular AI tools, who is responsible for prompts or inputs, who must validate generated material, when escalation is required, and who ultimately assumes responsibility for AI-assisted work.

Plan for failures before they occur. Vice Chancellor Will noted that the initial errors were compounded by counsel’s response and required an explanation of the remedial steps taken. Organizations should consider in advance how AI-related errors will be escalated, investigated, corrected, and documented. Leiske illustrates that a court evaluating AI-related failures may scrutinize not only the error itself, but also the adequacy of the organization’s response.

Leiske does not create new corporate AI-governance duties. But alongside An and American Council, the order reflects continued judicial attention to how organizations govern AI-assisted processes. As AI becomes embedded in more consequential workflows, courts may increasingly ask not only who made an error, but also what controls the organization had in place to prevent it.

This post is as of the posting date stated above. Sidley Austin LLP assumes no duty to update this post or post about any subsequent developments having a bearing on this post.