May 19, 2026
Five Things to Remember When Using AI for Business Decisions
The Bottom Line
Prompting an AI model can feel like creative brainstorming or thinking out loud. But in a court filing or displayed in a courtroom, an AI chat can seem like a confession. Here, the court found that Krafton “went searching for a pretext” to escape a payout it did not want to make, and its justifications for the firings were “manufactured” and “pretextual.”
If you are about to use AI to analyze a contractual obligation, personnel decision, or any other issue that could end up in court — pause first. Consider whether to consult your lawyer before putting your thoughts into an AI model. McGrath North’s Technology, Digital Infrastructure, Privacy and Cybersecurity, and Artificial Intelligence team is ready to help you analyze how to use AI in your business and when working with your legal team.
This Alert is for general informational purposes and does not constitute legal advice. Do not act or refrain from acting based on this Alert without seeking counsel regarding your specific circumstances. Attorney advertising.
Want To Use AI To Avoid Legal Obligations? Read This First.
A recent Delaware Chancery Court decision illustrates the risks of attempting to use AI to avoid a legal obligation. In Fortis Advisors, LLC v. Krafton, Inc., Krafton’s CEO wanted to avoid a contractual payout of up to $250 million involving the highly anticipated video game Subnautica 2. When his executive team and lawyers failed to suggest a way he could avoid the obligation, the CEO turned to ChatGPT. The AI chatbot initially suggested that it might be difficult to escape the obligation. Krafton’s CEO did not like that answer, so he kept prompting until the AI model produced a detailed “strategy” that he implemented. He deleted some of the chat logs, but the surviving records were introduced as trial exhibits — and were damning. In the court’s words, the CEO “consulted an artificial intelligence chatbot to contrive a corporate ‘takeover’ strategy.”
Why the case matters
The case shows how AI chats can become powerful evidence against a party. AI chats often involve a user repeatedly prompting until the AI model gives an answer the user wants. That sequence of prompts and answers can reveal the user’s thinking and motives. In addition, AI chatbots may generate answers that use provocative language. Here, for example, the AI model proposed a strategy it called a “pressure and leverage package” combining “hardball” and “softball” components. Those descriptions caught the court’s attention and were quoted extensively in the opinion. Finally, users may treat AI models as trusted advisors or even like therapists, creating a written record of thoughts and emotions that never would have existed before AI chatbots.
Dig Deeper: What Happened?
Krafton, a South Korean gaming conglomerate, acquired video game studio Unknown Worlds Entertainment in 2021 for $500 million, plus a potential earnout of up to $250 million tied to revenue targets. As the studio prepared to release Subnautica 2, it became clear to Krafton’s CEO that the new game was projected to trigger nearly the full earnout. Krafton's CEO concluded that he had signed a “bad deal” and would be branded a “pushover” if the earnout came due. Under the deal, the studio’s co-founders and its CEO, Edward Gill, retained operational control and could only be terminated for “cause,” a narrowly defined term in the deal documents. Krafton’s CEO consulted Krafton’s own executives and legal team, who warned that firing the studio’s leadership would not eliminate the earnout. ChatGPT initially agreed, telling him the earnout would be “difficult to cancel.” Undeterred, Krafton’s CEO kept prompting until the chatbot produced a detailed “pressure and leverage package” pairing “hardball (Legal + Finance)” tactics with “softball (Support / Incentives)” overtures.
To execute the plan, Krafton’s CEO formed an internal task force called “Project X.” It blocked Subnautica 2’s release, posted a negative message on the studio’s website addressed to its 12 million fans, and fired the studio’s co-founders and Gill, citing only their supposed intention to release the game prematurely.
After the sellers sued, the court reinstated Gill as studio head, restored his operational control, invalidated the board resolution blocking Subnautica 2’s launch, and extended the earnout window by 258 days to account for the wrongful conduct.
The court indicated that a second phase of the lawsuit may determine whether Krafton’s CEO’s actions impaired the earnout and what damages, if any, are owed.
Why the AI Record Was So Damaging
The AI chats revealed the CEO’s motivations and planning in real time, leading the court to conclude that Krafton’s supposed justifications for its actions were pretexts. Krafton could not outrun its own AI record.
Why the case matters
The case shows how AI chats can become powerful evidence against a party. AI chats often involve a user repeatedly prompting until the AI model gives an answer the user wants. That sequence of prompts and answers can reveal the user’s thinking and motives. In addition, AI chatbots may generate answers that use provocative language. Here, for example, the AI model proposed a strategy it called a “pressure and leverage package” combining “hardball” and “softball” components. Those descriptions caught the court’s attention and were quoted extensively in the opinion. Finally, users may treat AI models as trusted advisors or even like therapists, creating a written record of thoughts and emotions that never would have existed before AI chatbots.
Dig Deeper: What Happened?
Krafton, a South Korean gaming conglomerate, acquired video game studio Unknown Worlds Entertainment in 2021 for $500 million, plus a potential earnout of up to $250 million tied to revenue targets. As the studio prepared to release Subnautica 2, it became clear to Krafton’s CEO that the new game was projected to trigger nearly the full earnout. Krafton's CEO concluded that he had signed a “bad deal” and would be branded a “pushover” if the earnout came due. Under the deal, the studio’s co-founders and its CEO, Edward Gill, retained operational control and could only be terminated for “cause,” a narrowly defined term in the deal documents. Krafton’s CEO consulted Krafton’s own executives and legal team, who warned that firing the studio’s leadership would not eliminate the earnout. ChatGPT initially agreed, telling him the earnout would be “difficult to cancel.” Undeterred, Krafton’s CEO kept prompting until the chatbot produced a detailed “pressure and leverage package” pairing “hardball (Legal + Finance)” tactics with “softball (Support / Incentives)” overtures.
To execute the plan, Krafton’s CEO formed an internal task force called “Project X.” It blocked Subnautica 2’s release, posted a negative message on the studio’s website addressed to its 12 million fans, and fired the studio’s co-founders and Gill, citing only their supposed intention to release the game prematurely.
After the sellers sued, the court reinstated Gill as studio head, restored his operational control, invalidated the board resolution blocking Subnautica 2’s launch, and extended the earnout window by 258 days to account for the wrongful conduct.
The court indicated that a second phase of the lawsuit may determine whether Krafton’s CEO’s actions impaired the earnout and what damages, if any, are owed.
Why the AI Record Was So Damaging
The AI chats revealed the CEO’s motivations and planning in real time, leading the court to conclude that Krafton’s supposed justifications for its actions were pretexts. Krafton could not outrun its own AI record.
Five Things to Remember When Using AI for Business Decisions
- Your opponent may see your AI chats. AI conversations might be produced in discovery, just like email. That matters because your prompts may reveal your intent and motive, which are often the most critical issues in a dispute.
- Pushing past “no” can make things worse. If you keep re-prompting until the AI model gives you the answer you want, the entire sequence may end up in the court record. Then, the initial answer you rejected can become powerful evidence against you.
- Adopting an AI-generated plan turns the AI model’s words into yours. If your actions track the AI model’s suggestions, it is difficult to distance yourself from what the AI model said, even if you would not have originally chosen the words the AI model used.
- Deleting chats can backfire. Once a dispute is foreseeable, AI chat histories may be subject to the same preservation obligations as any other record. Deletions may be discoverable and might lead to court-imposed sanctions.
- Review your AI policy. An earlier McGrath North Alert discussed the risk that AI chats could expose otherwise-privileged information to an adversary. Your AI policy should account for that risk, and it also should provide broader guidance on how to use AI in your business.
The Bottom Line
Prompting an AI model can feel like creative brainstorming or thinking out loud. But in a court filing or displayed in a courtroom, an AI chat can seem like a confession. Here, the court found that Krafton “went searching for a pretext” to escape a payout it did not want to make, and its justifications for the firings were “manufactured” and “pretextual.”
If you are about to use AI to analyze a contractual obligation, personnel decision, or any other issue that could end up in court — pause first. Consider whether to consult your lawyer before putting your thoughts into an AI model. McGrath North’s Technology, Digital Infrastructure, Privacy and Cybersecurity, and Artificial Intelligence team is ready to help you analyze how to use AI in your business and when working with your legal team.
This Alert is for general informational purposes and does not constitute legal advice. Do not act or refrain from acting based on this Alert without seeking counsel regarding your specific circumstances. Attorney advertising.
Related Attorneys
