Earnouts, Bad Faith, and a Chatbot: What Fortis Advisors v. Krafton Means for Founders Who Sell

A $500 million deal with a $250 million catch

In 2021, the South Korean publisher Krafton acquired Unknown Worlds Entertainment, the studio behind the hit survival game Subnautica, for $500 million. That headline number was only part of the bargain. The contract layered a $250 million earnout on top of the purchase price — a contingent payout the founders could unlock if Subnautica 2 hit its sales targets after its 2025 Early Access release. In M&A terms, this is a familiar structure: a buyer pays a fixed amount at closing and promises a second, larger tranche only if the acquired business performs.

The earnout was not the only term that mattered. The equity purchase agreement (the EPA, the master contract governing the sale) guaranteed Unknown Worlds' independence and let the founders keep operational control of the studio. Cofounders Charlie Cleveland and Max McGuire, along with CEO Ted Gill, could be removed only “for cause.” That clause was the load-bearing wall of the whole deal. The people best positioned to hit the milestone were contractually entitled to stay in their seats and run the company through the earnout period — which is exactly what made the $250 million realistically earnable.

One more party belongs in the picture. The selling stockholders did not enforce the EPA individually. They acted through Fortis Advisors LLC, which served as the securityholder representative for the sellers — the single entity with standing to police the agreement after closing. When the relationship later soured, it was Fortis, not the founders personally, that brought suit in the Delaware Court of Chancery.

When hitting the target became the problem

The trouble was not that Subnautica 2 was failing. It was that the game was succeeding. In May 2025, Krafton's own finance department ran the numbers and projected a base-case earnout of roughly $191 million — a payout sitting close to the $250 million cap. By the company's internal math, the milestone was on track to trigger, and the nine-figure obligation that had sweetened the 2021 acquisition was about to come due.

That prospect did not sit well at the top. CEO Changhan Kim privately described the arrangement as a “pushover” contract, viewing the looming payment less as the price of a deal Krafton had agreed to and more as a financial liability — and, by his own framing, a threat to his standing. The earnout had done exactly what an earnout is supposed to do: it had paid off because the studio delivered. Inside Krafton, that outcome was treated as a problem to be solved.

The company was not blind to the risk of solving it the wrong way. Krafton's head of corporate development, Maria Park, warned that a “dismissal with cause” would not erase the $250 million obligation and would instead expose Krafton to lawsuit and reputation risk. In other words, leadership was told in advance that engineering a for-cause firing to dodge the earnout was both legally shaky and reputationally dangerous.

The Delaware Court of Chancery later reached the same conclusion about Krafton's intent. The court found that the company's true focus by June 2025 was avoiding its financial exposure — not any genuine worry about the game. Tellingly, Krafton's own CFO could not name a single employee who actually believed Subnautica 2 was unready for its planned release, which left the “quality and readiness” rationale looking manufactured rather than real.

'Project X': a CEO asks a chatbot how to escape a contract

Where most executives facing an inconvenient contract would call their lawyers, Changhan Kim opened a chat window. Bypassing Krafton's own legal team, Kim asked ChatGPT for advice on how to avoid paying the earnout. The chatbot's first answer was discouraging — it told him the obligation would be “difficult to cancel.” Kim kept prompting.

What he eventually pulled out was not legal advice but a strategy. ChatGPT produced a detailed multi-stage plan, which became an internal effort Krafton named “Project X”: an initiative aimed at either renegotiating the earnout or forcing a takeover of the studio. The advice, as later laid out in the litigation, read less like risk analysis and more like a script for manufacturing a justification after the decision had already been made. The plan told Kim to:

• form an internal task force to renegotiate the earnout or force a takeover of the studio;
• “lock down” Steam and console publishing rights, along with control of the game's code;
• frame the conflict as a fight over “fan trust” and “quality” rather than money;
• prepare systematic legal-defense materials while logging communications; and
• draft a public-facing message designed to win over fans.

Notice what that sequence does. The “fan trust” and “quality” framing supplies the pretext; the legal-defense materials build the file to defend it; the lockdown of publishing and code seizes the leverage to enforce it. This is the manufactured rationale from the prior section, assembled step by step. The internal warnings about the legal and reputational risk — the same concerns flagged inside the company — did not stop it; the founders were removed anyway, without legitimate cause. One detail foreshadows the trouble to come: the court later found that Kim had deleted specific relevant chat logs.

What the Delaware Court of Chancery actually held

On March 16, 2026, Vice Chancellor Lori W. Will of the Delaware Court of Chancery entered judgment for Fortis in Fortis Advisors, LLC v. Krafton, Inc. (C.A. No. 2025-0805-LWW). The court found that Krafton breached the equity purchase agreement by terminating the Key Employees without valid Cause and by improperly seizing operational control of Unknown Worlds. The firings, in other words, were not a real “for cause” removal at all.

The court treated the stated rationale as a cover story. Facing a nine-figure liability and frustrated by the founders' refusal to hand over control, Krafton — in the court's words — “went searching for a pretext.”

Krafton's defenses did not survive that framing. Its claim that the founders had threatened to self-publish failed. So did its reliance on after-the-fact allegations that an executive had improperly downloaded company data. Those downloads fell under the after-acquired evidence doctrine — misconduct a company discovers only after it fires someone, which generally cannot retroactively justify a termination decided for other reasons. The court required Krafton to show both that the conduct met the contractual standard and would have independently prompted termination, and the defensive data-copying did not clear that bar. A related principle, the “mend-the-hold” doctrine, barred Krafton from inventing new justifications it had known about but not invoked when it acted.

The remedy is what makes the case unusual. Rather than awarding money, the court ordered specific performance — a court order forcing a party to actually do what it promised. CEO Ted Gill was reinstated with restored Steam access and authority over the Subnautica 2 Early Access launch, and the July 1, 2025 board resolution was set aside to the extent it infringed his operational-control right. The court also extended the earnout Testing Period by 258 days — the length of Gill's ouster — to September 15, 2026, with Fortis keeping a contractual right to push it to March 15, 2027. That is an equitable extension: the clock is reset so the founders get the time the breach took from them.

Read this for what it is. It is a major interim resolution, not the final word. Krafton said it “respectfully disagree[s]” and is weighing its options, including a possible appeal, and the ruling does not resolve the former executives' damages claim or the earnout itself. Further litigation is still pending.

The doctrine: earnouts, 'efforts' clauses, and the implied covenant of good faith

Krafton's conduct was extreme, but the dispute it produced is ordinary. Earnouts exist to bridge a valuation gap: the buyer will not pay today for performance it cannot yet see, so it promises to pay later if the target hits agreed milestones. That structure quietly shifts performance risk onto the seller and hands the buyer the keys to the business that has to perform. The result is a predictable conflict — buyers want operational discretion, sellers want assurance the milestone will actually be pursued — and Delaware practitioners describe it as an all-too-predictable pattern of post-closing litigation over whether the targets were met.

Delaware polices the buyer's post-closing conduct in two ways. Where the contract contains an express efforts covenant — “commercially reasonable efforts,” “best efforts,” or similar — the court enforces that language; Delaware treats those variations as roughly interchangeable standards and reads the agreement's plain terms to fix how much discretion the buyer keeps. Where the contract is silent, the implied covenant of good faith and fair dealing may apply, but only narrowly. Delaware calls it a limited and extraordinary remedy that fills a genuine gap for a truly unanticipated development — it cannot rewrite a risk the parties could have addressed at the bargaining table.

The two leading guideposts cut both ways. In the STX line of cases, a buyer with bargained-for discretion does not act in bad faith merely by using it — but that protection runs only “so long as the buyer did not take action in bad faith or with the specific intention of causing a reduction in the Earnout.” And in Johnson & Johnson v. Fortis Advisors LLC (Del. 2026) — a separate matter that happens to involve the same repeat-player seller representative, not the same deal — the Delaware Supreme Court read a commercially-reasonable-efforts standard as “inward-facing” and cabined: the buyer could weigh enumerated business factors in calibrating effort, but could not use them to undermine the milestones. Notably, the seller did not prevail on the first milestone, a reminder that these cases do not reflexively favor sellers. For founders fighting over wording, the lesson is that the precise label on an efforts clause matters less than securing an express covenant at all.

Krafton sits on the wrong side of that line, and that is what makes it stand out. Most earnout plaintiffs cannot prove the buyer's motive, so they settle for hard-to-win damages. Here the seller could show Krafton acted with the specific intention of avoiding the milestone as projections approached the $250 million cap — and the court responded by enforcing the bargained-for control provisions directly, reinstating the CEO and extending the earnout period rather than awarding money. Specific intent plus equitable relief is the uncommon combination; the doctrine it applied is not.

The AI-governance lesson: don't outsource good faith to a chatbot

Strip away the gaming-industry color and the chatbot transcript at the center of this case is something every executive now has within reach: a fast, fluent tool that will happily generate a plan when you ask it for one. The problem in Krafton was never that an AI was used. It was that a consequential, legally fraught decision — how to escape a $250 million obligation — was steered by a chatbot session instead of independent human judgment exercised through counsel.

Vice Chancellor Will drew that line directly. As Fortune reported, the court noted that company executives are expected to exercise independent human judgment — not to outsource good-faith decisions to an AI. Read in context, that is not a pronouncement about AI law. It is a good-faith holding in which the AI conduct served as evidence of intent. “The AI told me to” is not a defense; an AI-generated plan to manufacture cause reads closer to an admission of the very state of mind the implied covenant forbids.

Two practical points follow, both familiar from how courts treat email and Slack. First, generative-AI prompts and outputs are discoverable. The transcript documented the improper motive in the company's own words, and practitioners reviewing the case have warned that AI-assisted strategic planning creates discovery risk and can expose improper motivations once it is written down. Second, deleting the logs does not make them disappear from the case. As a general matter, destroying relevant records while litigation is anticipated invites spoliation sanctions and adverse-inference instructions — and the attempt to hide the AI use only deepens the credibility problem.

None of this argues for technophobia. The lesson is that AI belongs inside a governed process, supervised by human judgment and counsel. For founders and in-house teams structuring sensitive decisions, that means building AI governance into how the company actually operates:

• Route consequential legal or strategic questions through counsel. Lawyer advice can carry privilege; a chatbot session does not.
• Treat AI as a drafting and research aid, not a decision-maker. A human owns the call and the rationale behind it.
• Assume every prompt and output may be read aloud in court, and preserve records rather than purge them.

If you are the founder selling: build the protections into the deal

The most useful sentence for sellers is one the Delaware Supreme Court has applied to earnout fights elsewhere: the implied covenant of good faith “cannot be invoked to provide protections that ‘easily could have been drafted’ at the bargaining table.” Read that as an instruction. Almost everything the Unknown Worlds sellers won, they won because the right language was already in the equity purchase agreement — and almost everything still in dispute is something the contract could have nailed down but didn't. Translate the ruling into deal terms before you sign.

• Define “Cause” tightly and require real process. Delaware reads narrow “for cause” definitions strictly and demands proof of conscious wrongdoing, not merely unauthorized conduct, so a vague standard invites a manufactured pretext. Spell out the grounds, add notice and a cure period, and make sure the earnout mechanism accounts for a founder's departure rather than leaving termination as the buyer's escape hatch. Tie your operational control directly to the earnout period — that linkage gave the founders something a court could restore.
• Negotiate an affirmative efforts covenant and objective milestones. Put a “commercially reasonable efforts” (or more specific) obligation on the buyer in writing, define the milestone in measurable terms, and add reporting and audit rights so you can see what the buyer is doing. Where the contract is silent, you are left arguing the implied covenant, which only fills genuine, unanticipated gaps and cannot rewrite a deal you could have drafted differently.
• Add anti-circumvention and specific-performance language. An express equitable-relief clause is what let the court reinstate the CEO “with full operational authority” and extend the earnout period “by the length of his wrongful ouster” rather than send the sellers off to prove speculative damages. The right to compel restored control is worth far more than a hard-to-quantify damages claim.
• Appoint a securityholder representative. Fortis Advisors — not the founders personally — brought and prosecuted the claims on the sellers' behalf. Individual founders rarely have the standing or resources to litigate a post-closing earnout alone; a dedicated representative gives the seller side a party that can actually enforce the deal.
• Consider acceleration or “deemed-earned” triggers. A provision that treats the milestone as satisfied if the buyer removes key leadership, blocks the launch, or otherwise interferes converts a contested factual fight into a clean contractual trigger.

None of this is one-size-fits-all. The right “Cause” definition, efforts standard, and remedy package depend on your leverage, the asset, and the governing law, so treat this as educational rather than advice on your specific transaction. The through-line is simple: the protections you can write down beat the good faith you would otherwise have to litigate.

Key implications for practice

Three lessons survive the noise of this case, one for each side of a deal table.

If you are selling, your earnout is only as valuable as the protections around it. Delaware enforces what the contract says and uses the implied covenant only to fill genuine gaps. The headline number means little without tight control terms, a real efforts covenant, and a remedy a court can actually enforce. Negotiate those at signing, because you will not get them later.

If you are buying, or sitting in the in-house seat, your post-closing conduct is judged on intent. The court found that Krafton, facing nine-figure liability, “went searching for a pretext.” Manufacturing a reason to dodge an earnout proved more expensive than honoring it: instead of paying, the buyer lost control of the studio and the timeline. Pretext is not a strategy. It is evidence.

For everyone, AI belongs inside a governed legal process, not in place of one. As the court observed, executives are expected to exercise independent human judgment — not to outsource good-faith decisions to an AI. Used within counsel's process, these tools help. Used to engineer a way out of a promise, they leave a transcript.

One caveat worth repeating: this is an interim resolution. Damages and the earnout itself remain to be litigated, and Krafton has signaled it may pursue other options. Treat the ruling as a clear signal, not the last word.