A Billion-Dollar CEO Asked ChatGPT How to Steal $250 Million From His Own Studio. A Delaware Judge Said No.

There are bad ideas, there are catastrophic ideas, and then there is asking a chatbot for a strategy to strip $250 million in contractually owed money from the founders of a studio that trusted you enough to sell themselves to your company. That last category now has its defining example, and it is sitting in the Delaware Court of Chancery docket under the caption Unknown Worlds Entertainment v. Krafton.

On March 16, 2026, Vice Chancellor Lori W. Will of the Delaware Chancery Court issued a ruling that should be required reading for every executive currently treating ChatGPT as a free substitute for legal counsel. She found that Krafton, the South Korean publisher behind PUBG, had breached its purchase agreement with Unknown Worlds Entertainment, the developer of Subnautica. She ordered the ousted CEO reinstated. She extended the deadline for a $250 million earnout that Krafton had been quietly maneuvering to avoid paying. And she did it all after reviewing court filings that laid out, in excruciating detail, how Krafton CEO Changhan Kim had used ChatGPT to build the entire takeover strategy.

This is not a story about ChatGPT getting a fact wrong. It is a story about a billion-dollar executive deliberately feeding the chatbot a question nobody in his own legal department would answer the way he wanted, getting the answer he wanted, and then attempting to execute it in the real world. The AI did exactly what AI does. It generated plausible-sounding text. The humans who then turned that text into corporate action are the ones now eating a court order.

The $500 Million Deal That Created a $250 Million Landmine

To understand how stupid this gets, you need to understand the original transaction. In 2021, Krafton bought Unknown Worlds Entertainment for $500 million. The studio, founded by Charlie Cleveland and Max McGuire with Ted Gill running operations as CEO, was coming off the massive success of the original Subnautica, a survival game that turned into a cultural phenomenon.

Krafton wanted the franchise, but it also wanted to keep the people who made it work. So the Equity Purchase Agreement included an extra incentive: up to $250 million in additional earnout payments, contingent on the performance of Subnautica 2. Hit the targets, collect the bonus. Miss them, no bonus. Standard stuff in gaming M&A deals, designed to align the incentives of the acquiring company with the creative people who actually build the product.

The agreement also had a very standard protection for the sellers. Unknown Worlds was supposed to remain operationally independent, and the three key employees, Cleveland, McGuire, and Gill, could only be removed for cause. Translation: Krafton could not simply fire them, claim the studio was underperforming, and walk away from the $250 million obligation. This is the kind of contractual guardrail every gaming acquisition lawyer puts in place precisely because the alternative is obvious.

Then Subnautica 2 started looking like a hit. Krafton's own internal sales projections showed the game was trending toward the performance thresholds that would trigger the full $250 million payout. Which is exactly the scenario the earnout was designed to reward. For a normal company, this is a champagne moment. You bought the right studio, they are delivering a hit, and you pay the bonus because you are now sitting on a franchise worth many multiples of that number.

Krafton, apparently, did not feel celebratory.

When Your Own Lawyers Say No, You Ask ChatGPT

According to the court filings that eventually became public through the Delaware proceedings, Changhan Kim asked his internal legal team a straightforward question: could they fire the three key employees for cause and escape the earnout? The internal lawyers gave him the answer that lawyers give when the facts do not support the outcome a client wants. They told him that a dismissal with cause would not eliminate the obligation. The contract was written carefully. The earnout was protected.

Kim did not accept that answer. According to the court's findings, he went to ChatGPT instead.

The prompts became part of the record. Kim asked the chatbot whether the earnout could be canceled. ChatGPT, to its very limited credit, initially responded that the earnout would be "difficult to cancel." A competent lawyer reading that output would close the tab and go to lunch. Kim pushed further. He reframed the question, asked about alternative corporate structures, asked what a CEO in his position could theoretically do. And ChatGPT, because ChatGPT is a text generator that produces plausible continuations rather than legal advice rooted in fiduciary duty, obliged.

The chatbot laid out a detailed, multi-stage corporate strategy. It advised Kim to create an internal task force with a mandate to either renegotiate the earnout or execute a takeover of Unknown Worlds. It suggested that if the negotiations failed, Krafton should "lock down" Steam and console publishing rights, and seize control of the game's source code. It recommended framing the entire conflict publicly as a dispute about "fan trust" and "quality," rather than the actual dispute, which was about $250 million. It told Kim to start logging all communications and preparing legal defense materials in advance, so that when the lawsuit eventually came, Krafton could argue it had been acting in good faith the whole time.

Read that again. A chatbot, asked how to accomplish something the asker's own lawyers refused to help with, produced a step-by-step playbook for a hostile takeover of an independent studio, complete with PR strategy and pre-emptive legal narrative construction. And a Fortune 500 CEO executed it.

Project X: The AI-Drafted Corporate Takeover

Internally, Krafton named the initiative Project X. The task force did exactly what the chatbot suggested. Krafton began publicly criticizing the quality and direction of Subnautica 2, setting up the "this is about fan trust" narrative. The company moved to oust Ted Gill as CEO of Unknown Worlds. It cleared the path to remove Cleveland and McGuire from operational roles. And it moved to take direct control of the publishing, distribution, and development pipelines, cutting the original leadership out of the decisions that determined whether the earnout thresholds would be hit.

If you are reading this and thinking that it sounds exactly like the breach of contract the Equity Purchase Agreement was designed to prevent, congratulations, you are smarter than the strategy the CEO of a publicly traded gaming company decided to pursue. The court agreed with you. Vice Chancellor Will concluded that "Krafton breached the EPA by terminating the Key Employees without valid Cause and by improperly seizing operational control of Unknown Worlds."

The task force's own documents, subpoenaed during discovery, were so damning that the opinion essentially wrote itself. It is very difficult to argue you had legitimate performance concerns about a studio when your own internal communications describe a strategy designed to prevent a bonus from triggering. It is even harder when that strategy came from a source that every grown-up lawyer has been warning executives to stop using for legal advice since at least 2023.

The Takeover Playbook the Court Saw

Based on the court's findings and the filings reported by outlets covering the case, the ChatGPT-generated strategy included:

1. Form an internal task force to either renegotiate the earnout or execute a takeover.
2. If negotiation fails, lock down Steam and console publishing rights.
3. Seize control of the game's source code and development pipeline.
4. Frame the public conflict as being about quality and fans, not money.
5. Prepare legal defense materials and log every communication proactively.
6. Remove the key employees and justify it with manufactured performance concerns.

Every one of those steps showed up in Krafton's actual conduct. The court did not have to speculate about parallels. The parallels were the case.

The Delaware Chancery Court Ruling

Delaware Chancery Court is where the world's corporate disputes get decided, and Vice Chancellor Lori W. Will has been on that bench long enough to recognize a bad-faith takeover when one lands on her docket. On March 16, 2026, she ruled in favor of Unknown Worlds, reversing virtually every move Krafton had made.

Ted Gill was ordered reinstated as CEO of Unknown Worlds. He was given authority to bring Cleveland and McGuire back into their operational roles. Krafton was enjoined from interfering with Unknown Worlds' management of Subnautica 2's early access launch. And the court extended the earnout window to account for the disruption Krafton had caused, meaning the cofounders are now eligible to collect the full $250 million through September 15, 2026.

If you were designing a lesson about how not to use generative AI in high-stakes corporate decisions, you could not write a cleaner one. The plan was generated by a chatbot that had no knowledge of Delaware contract law, no understanding of the specific EPA Krafton had signed, and no fiduciary relationship with anyone involved. It generated a plausible-sounding sequence of steps because that is what language models do. When those steps were executed in the real world by a CEO who should have known better, they collided with the actual law, and the actual law won.

This Was Not a Hallucination. It Was Something Worse.

The easy take on this story is that ChatGPT hallucinated, gave bad legal advice, and the CEO was unlucky. That take is wrong, and it lets the CEO off the hook in a way that should bother anyone paying attention to how AI is being integrated into corporate decision-making.

ChatGPT did not hallucinate a legal principle here. It did not cite a made-up case. It did not misstate a Delaware statute. What it did was something more insidious: it produced a plausible strategic plan that was internally consistent, sounded sophisticated, and happened to describe a course of action that was unambiguously illegal under the specific contract in question. The chatbot was not pretending to be a lawyer. The CEO was pretending it was a lawyer. The responsibility for the failure sits on the human who took the text output of a statistical model and turned it into a corporate directive.

This is the uncomfortable truth about the current generation of AI tools. Their most dangerous output is not the obvious hallucination. It is the confident, polished, professional-sounding advice that is exactly wrong for your specific situation in ways only a human expert could recognize. The CEO asked a question that required understanding of the specific EPA, the specific Delaware fiduciary duty doctrine, and the specific facts of the Subnautica 2 development. ChatGPT cannot know any of that. It can only generate text that sounds like something a consultant might say.

That is why every serious profession has been screaming about this since 2023. Lawyers. Doctors. Accountants. Financial advisors. The warning has always been the same: these tools are excellent at producing first drafts of generic content, and catastrophic at producing bespoke advice that hinges on facts and expertise they cannot access. Krafton's CEO did not listen. The Delaware court made him listen.

The Pattern: Executives Are Outsourcing Judgment to a Chatbot

The Krafton case is the most expensive example of a pattern that has been building for two years. A lawyer in New York gets sanctioned for filing a brief with ChatGPT-hallucinated case citations. A federal appeals court fines another attorney $2,500 in early 2026 for AI-generated legal fabrications. Academic papers are getting retracted because researchers pasted ChatGPT output directly into citations. The failures keep arriving at different levels of the professional stack, and the common thread is always the same: a person with real authority outsourced a high-stakes judgment call to a text generator.

What makes Krafton different is the scale of the money, the seniority of the person involved, and the deliberate nature of the decision. This was not a lazy associate trying to save time on a brief. This was a CEO actively routing around his own internal legal team because they refused to help him do something he wanted to do. ChatGPT did not fail him. ChatGPT enabled him. It was the compliance-free second opinion he was looking for, and it gave him exactly the hallucinated confidence he needed to proceed with a plan his own lawyers had already flagged as problematic.

That is the real disaster. Not that the AI was wrong. That it will always, always tell you something. A human lawyer with a license to protect and a duty to the client will sometimes refuse to answer a question because the answer is no. A chatbot will never refuse. It will always generate a plan. And for an executive who has decided in advance what he wants to do, that unconditional willingness to produce text is not a feature. It is a trap.

Timeline: How Krafton Walked Into the Courtroom

There is a version of this story where AI makes executives smarter. Where it synthesizes complex legal and strategic information, flags risks humans miss, and helps leaders make better decisions. That version requires the human to treat the AI as a tool that informs their judgment rather than replaces it. It requires the human to have the judgment in the first place.

Changhan Kim did not have that judgment. He had an outcome he wanted, a legal team that told him the outcome was not available, and a chatbot that would tell him whatever he typed a prompt clever enough to extract. When the Delaware Chancery Court finally looked at what happened, it saw exactly what anyone with a functioning sense of corporate governance would have seen from day one. A contract. A bad-faith attempt to evade it. And an AI-generated trail of evidence so clean it practically convicted him by itself.

The Krafton case will be cited for years. Business school case studies will use it. Corporate compliance departments will circulate it. Every internal AI governance policy written in the back half of 2026 will include some version of the same rule: do not use a chatbot to route around your own legal team. The lesson cost Krafton a quarter of a billion dollars. For everyone else, it is free. Use it.