What do an underwater survival exploration game set on an alien planet, a corporate acquisition, ChatGPT, and a state court in Delaware have in common? They came together in a breach of contract bench trial to shine a spotlight on the potential discoverability of AI-generated data and documents. For organizations using generative AI, this case offers a stark warning: what employees ask AI tools, and what those tools generate in response, may later become some of the most damaging evidence in litigation.

Background

In March 2026, the Delaware Court of Chancery issued its decision in Fortis Advisors, LLC v. Krafton, Inc., C.A. No. 2025-0805-LWW, finding that the defendant company breached an equity purchase agreement (EPA) by terminating key executives without cause and seizing operational control of the studio to avoid paying a contractual earnout of up to $250 million.

The decision arises out of a disputed contract entered into in 2021, when South Korean gaming company Krafton acquired Unknown Worlds Entertainment—the studio behind the underwater survival game Subnautica—for $500 million upfront plus an earnout of up to $250 million tied to studio revenue through 2025. The deal guaranteed that Unknown Worlds’ founders and CEO would retain operational control and could be terminated only for "Cause" as narrowly defined in the EPA.

As Unknown Worlds prepared to launch Subnautica 2 in August 2025, Krafton's finance team projected that the release would trigger a nine-figure earnout payment. Rather than honor the contract, Krafton's CEO consulted ChatGPT for guidance on avoiding the payment. The AI chatbot produced a detailed "Response Strategy to a 'No-Deal' Scenario," complete with steps for locking down the studio's publishing platform, preparing legal defense materials, and executing a corporate takeover. Krafton’s CEO followed most of the ChatGPT-generated playbook: it seized the Steam publishing platform, posted unauthorized messages on the studio's website, and on July 1, 2025, terminated Unknown Worlds’ founders and CEO on a pretext of game unreadiness. Krafton’s CEO later admitted to deleting the relevant ChatGPT logs.

The Court found that Krafton’s terminations were pretextual and breached the EPA. It reinstated the Unknown Worlds’ CEO with full operational authority, enjoined Krafton from circumventing his contractual rights, and extended the earnout period by 258 days to make up for the time lost to Krafton's breach. The AI-generated strategy document, which was produced in discovery, was quoted at length in the opinion as direct evidence that Krafton's justifications were manufactured after the fact.

What the Case Reveals About AI and Discovery

The Fortis Advisors case illustrates a risk that many organizations have not yet fully internalized: AI-generated content—and the prompts used to generate it—created in the ordinary course of business and retained, is a corporate record. It is subject to litigation holds, preservation obligations, and discovery. These business records, which are often unfiltered and revealing, may be used in litigation as evidence of an organization’s state of mind and knowledge.

In this case, the ChatGPT strategy document did not merely corroborate other evidence of bad faith. It was the evidence. The document provided a point-by-point roadmap of Krafton's actual intentions, against which the Court measured every subsequent action and every after-the-fact litigation justification. The CEO's decision to delete the ChatGPT logs only compounded the problem, inviting the Court to consider adverse inferences.

The Krafton case offers a real-world, in-the-courtroom perspective on a reality – employee usage of AI can generate data and documents that may be discoverable. That risk is compounded by the likelihood that employees will turn to AI to help with problems involving disputes, compliance, personnel decisions, contractual obligations, and other areas involving litigation and regulatory risks.

Practical Next Steps for Organizations

We can expect that courts will continue to grapple with the discoverability of data and documents generated by organizational use of AI. It would be prudent for most organizations to consider taking the following next steps:

1. Inventory Your Organization’s AI Use

Most organizations do not have a complete picture of which AI tools their employees are using, for what purposes, or what data those tools are retaining. These gaps can create legal risks. Organizations should inventory employees’ use of AI, assessing these questions and possibly others:

• Which AI platforms are in use across business units, including both enterprise-licensed tools and consumer applications accessed through personal accounts.
• Whether any current AI use touches areas of legal sensitivity—active or anticipated disputes, regulatory matters, confidential business information, significant negotiations, personnel decisions, or contract performance questions.
• Whether existing acceptable-use, data classification, and records retention policies expressly address AI-generated content, and whether those policies are being followed.

2. Update Litigation Hold Memoranda to Cover AI-Generated Data

Standard litigation hold notices predate the widespread adoption of generative AI and do not capture the relevant data categories. When a hold is triggered, whether by a filed complaint, a threatened claim, or a reasonably anticipated dispute, the notice should expressly require preservation of specific categories of information, including:

• AI platform chat histories and session logs. This material should include conversations stored in browser-based dashboards, exported files, or application data.
• Certain prompts and responses. Preserve if the subject matter relates to the dispute, the relevant transaction, the personnel involved, or any contractual obligation at issue.
• AI-generated drafts, analyses, summaries, and strategic documents. These materials, if shared with and/or acted upon by decision-makers, should be preserved.
• Platform auto-delete settings. Many AI tools purge conversation histories after a set period. Hold notices should require employees to disable auto-delete features and preserve existing logs immediately upon receipt of the hold.

3. Treat AI-Generated Content as Discoverable – and Plan for Privilege Review

AI-generated content used in business decision-making is generally not protected by the attorney-client privilege or the work product doctrine. Privilege attaches to confidential communications made for the purpose of obtaining legal advice, from or to an attorney. As demonstrated in the recent decision in U.S. v. Heppner, AI tools are not attorneys and consulting with them independently, may wave attorney-client privilege. Accordingly, a business executive’s use of an AI tool to evaluate strategic options—without the direction of counsel—is not privileged and is likely discoverable.

Organizations should work with in-house counsel to create rules for different types of AI usage, including:

• AI use in ordinary business operations. This material is likely discoverable as a business record regardless of later legal review;
• Define limits and conditions of AI usage. Consider allowing only enterprise-grade AI tools for any tasks related to legal, regulatory, compliance, or sensitive business issues, preferably at the direction of counsel; and
• AI use directed by counsel as part of the legal representation of the organization. This material should be protected by attorney-client privilege and/or work product doctrine if properly prepared and maintained.

Employees engaged in sensitive matters, such as active disputes, regulatory inquiries, significant contract negotiations, or personnel actions, should be required to consult with counsel before using AI tools for strategic guidance.

Conclusion

Fortis Advisors v. Krafton will not be the last case in which AI-generated content plays a decisive role at trial. As generative AI becomes further embedded in corporate decision-making, the volume of potentially discoverable AI records will continue to grow. Organizations that treat AI interactions as ephemeral are underestimating significant and growing litigation and regulatory risks.

For More Information

Van Ness Feldman’s litigators closely monitor and advise clients on AI-related issues. For more information, please contact Michael Farber, Justin Panitchpakdi, or any member of VNF’s litigation team.