Jones v. Delta Air Lines, Inc., No. 2:24-cv-11224 (E.D. Mich. Apr. 22, 2026)

Why This Case Is Important

This landmark ruling confirms that organizations cannot offload liability to generative AI. When integrating AI into core workflows, courts will scrutinize corporate governance, human oversight, and verification processes rather than treating the technology as an independent actor.

Overview Text

In American Council of Learned Societies v. National Endowment for the Humanities, a federal judge ordered the restoration of vital funding after a historical organization challenged the constitutionality of the termination of more than 1,400 federal grants. The dispute arose after government personnel delegated a high-stakes screening mechanism to an LLM (Large Language Model), specifically ChatGPT, to identify programs that implicated Diversity, Equity, and Inclusion (DEI) criteria.

Personnel entered grant descriptions into the platform using a rigid prompt: “Does the following relate at all to DEI? Respond factually in less than 120 characters. Begin with 'Yes.' or ‘No.’” The resulting classifications were embedded into the final decision-making framework without meaningful human review, prompting a lawsuit over algorithmic bias and deficient oversight. The government attempted to evade constitutional liability by pointing to errors in the software’s output. Displeased with this defense, the Southern District of New York methodically evaluated the limits of automated accountability.

Ruling Summary

• No Escaping Liability via Algorithmic Scapegoating The court forcefully rejected the government's attempt to deflect blame onto its automation vendor, writing that a party “cannot escape liability… by scapegoating ChatGPT”. Because the LLM served as the organization’s “chosen instrument for purposes of this project,” the corporate entity remains fully legally responsible for any defective or unlawful outcomes generated by the system. Alluding to classic pop-culture excuses, the court famously noted: “The devil made me do it. That excuse did not work for Geraldine Jones, and it does not work for the Government”.
• Nominal Human Involvement Declared Insufficient The court determined that perfunctory or uncritical approval of algorithmic outputs is insufficient for substantive corporate governance. A critical failing was the finding that no instances were recorded where a human reviewed a system-produced finding, exercised judgement to modify the decision, or chose a different course of action. Legal protection fails if a human-in-the-loop workflow is nominal, passive, or fundamentally incapable of identifying contextual errors in automated text.
• Prompt Engineering Established as a Governance Duty The court determined that the specific way instructions are engineered into the system has a direct bearing on a company's regulatory adherence. It noted that the creation of prompts serves as a critical governance factor, especially when employees without necessary subject-matter expertise develop queries that trigger automated results without accounting for essential background details. Organizations must maintain strict administrative controls over who is authorized to draft operational prompts for automated workflows.

Expert Analysis

Nancy Patton, Esq., CEDS, Senior Director, Solutions Engineering, Exterro

Both plaintiff and defendant motioned for Summary Judgment in this case and the plaintiff prevailed.This ruling by the SDNY means that the judge believes there are no genuine disputes of material fact as presented by the plaintiff. What is notable about that is the unprecedented nature of this case in terms of AI decision making in the context of organizational oversight and yet there is NO dispute. In the rapidly evolving and inevitable use of AI in organizations, the indisputable truth is that humans must understand and monitor AI. AI prompting cannot be a guess. AI outputs cannot be unsupervised. AI decisions cannot be autonomous. AI cannot ignore human involvement. The lawsuits that spawn from the use of AI that does not involve human oversight are barely getting started. Keep an eye out; each one is a fascinating evolution of our legal jurisprudence on this topic and provides a clear cautionary tale of what not to do.

Tip Text

Treat generative prompts, model parameters, and pilot system evaluations as highly discoverable data points in modern litigation. You cannot rely on a generic human sign-off; instead, execute documented validation testing, escalation mandates, and structured accountability protocols. Courts will mandate exhaustive process logs when AI dictates operational business decisions.

‍This alert is for informational purposes only and is not legal advice.

‍

‍

‍

‍

‍

‍