Krafton CEO Used ChatGPT for Legal Advice—Lost Big

What’s Happening

In what may become the most cautionary tale of 2026 so far, Chang Byung-gyu, CEO of South Korean gaming giant Krafton—the company behind the massively popular PUBG franchise—reportedly turned to ChatGPT for advice on how to void a $250 million contract. According to reporting from Reuters and coverage across Korean business media, Chang fed contract details into OpenAI’s chatbot, received what he apparently considered a viable legal strategy, and then proceeded to act on that strategy—over the explicit objections of his own legal team.

The result? A spectacular loss in court. The presiding judge reportedly found Krafton’s arguments not only unpersuasive but procedurally flawed in ways that experienced contract attorneys would have immediately flagged. The counterparty was awarded damages, and Krafton now faces an even steeper financial and reputational hit than the original deal would have cost them. The case has ignited a fierce debate in corporate governance circles about AI reliance in high-stakes decision-making, and it’s sending shockwaves through the startup world, where founders routinely use ChatGPT and similar tools for everything from pitch decks to term sheet analysis.

The story went viral this week after court documents became partially public and Korean tech journalists began connecting the dots. As The Verge noted, this isn’t the first time AI-generated legal content has backfired in court—recall the 2023 case where a New York lawyer cited fabricated cases hallucinated by ChatGPT—but it is by far the highest-dollar example of a corporate executive deliberately choosing an AI chatbot’s advice over qualified human counsel.

Why Founders Should Care

Let’s be honest: if you’re a startup founder in 2026, you’ve probably used ChatGPT or Claude or Gemini to draft a contract clause, interpret a term sheet, or figure out whether that non-compete your last employer made you sign is actually enforceable. There’s nothing inherently wrong with that. AI legal tools are genuinely useful for first-pass analysis, brainstorming, and even drafting routine documents. The problem is when you mistake a language model’s confident-sounding output for actual legal expertise—and when you let it override the humans you’re paying to protect you.

This matters now because we’re at an inflection point. The AI-powered legaltech market is projected to hit $2.6 billion by 2027, according to Bloomberg Law research. Startups are building real products in this space—contract review tools, compliance automation platforms, AI-assisted due diligence. Investors are pouring money in. But the Krafton debacle just handed every skeptic a giant megaphone. Regulators who were already eyeing AI-in-law guardrails now have a nine-figure case study to point to. If you’re building in legaltech, your go-to-market narrative just got more complicated. And if you’re a founder relying on AI for legal decisions, you just got a very expensive reminder of where the boundaries are.

The timing signal here is unmistakable: the market for AI-assisted legal work is massive and growing, but the gap between “AI-assisted” and “AI-decided” is where careers and companies go to die. That gap is your opportunity—and your risk.

Who’s Already Moving

Several startups and established players are building in the legaltech AI space, and the Krafton case is going to reshape how they all position themselves:

• Harvey AI — Backed by Sequoia and built on OpenAI’s models, Harvey has raised over $100 million and works with elite law firms like Allen & Overy. Their pitch has always been “AI for lawyers, not instead of lawyers,” and you can bet they’re going to lean even harder into that framing now. Harvey’s tool is designed to augment attorney workflows—research, drafting, summarization—not replace legal judgment. TechCrunch reported on their Series B in late 2024, and growth has been strong since.
• Ironclad — This contract lifecycle management platform has been one of the breakout enterprise legaltech companies of the past few years. They use AI to streamline contract creation and approval workflows but keep humans in the loop for review and sign-off. Post-Krafton, their “human-in-the-loop” design philosophy looks prescient.
• Spellbook — A Canadian startup that integrates directly into Microsoft Word to assist lawyers in drafting and reviewing contracts. They’ve been explicit about positioning their AI as a “co-pilot, not an autopilot”—a distinction that now carries real weight in sales conversations.
• EvenUp — Focused on personal injury law, EvenUp uses AI to generate demand letters and case analyses. They’ve been growing rapidly but have also invested heavily in quality control layers, understanding that one bad AI-generated legal document could tank the company.

“AI can tell you what a contract says. It cannot tell you what a contract means in the context of your specific business relationships, regulatory environment, and strategic goals. That requires judgment. And judgment is still a human product.” — Marc Lauritsen, Harvard Law School lecturer and legal technology researcher

The Startup Opportunity

The Krafton disaster doesn’t kill the AI-legal space. It sharpens it. Here’s where founders should be looking:

• AI guardrails for executive decision-making. There’s a product to be built that sits between a CEO and an AI output—essentially a “confidence scoring” layer that flags when an AI-generated legal or strategic recommendation is likely unreliable, hallucinated, or dangerously oversimplified. Think of it as a “second opinion” engine. The market for this is every company where non-lawyers are using AI for legal-adjacent decisions, which is… basically every company.
• AI-powered legal triage for startups. Most early-stage founders can’t afford outside counsel for every question. A tool that helps founders understand when they need a real lawyer—and when a templated AI-assisted answer is probably fine—would be enormously valuable. The key differentiator: being honest about its own limitations. No one’s really nailed this yet.
• Compliance and audit trails for AI-assisted decisions. After this case, corporate boards and investors are going to start asking: “How are you using AI in decision-making, and what’s the oversight process?” A startup that builds compliance tooling around AI usage in enterprise settings—logging what prompts were used, what outputs were generated, who approved what—is sitting on a gold mine. This is governance infrastructure, and it barely exists today.
• Training and certification for “AI-literate” executives. Not a SaaS play exactly, but there’s a B2B education opportunity here. Enterprises need their leaders to understand what AI can and can’t do. The Krafton case is going to be in every corporate risk management presentation for the next five years. Someone’s going to build the training platform that capitalizes on that.

Watch Out For

This story is a minefield of cautionary signals. Here’s what could go wrong for founders trying to build or operate in this space:

• Regulatory backlash could freeze the market. Governments in the EU, South Korea, and the US are already debating AI regulation in professional services. A high-profile failure like this gives regulators ammunition to impose strict licensing or certification requirements on AI legal tools. If you’re building in this space, you need to be tracking regulatory developments in every target market—weekly, not quarterly.
• Liability is still an open question. If your AI tool gives bad legal advice and a customer acts on it, who’s liable? You? OpenAI? The customer? The answer is: nobody really knows yet, and the case law is still being written. This is existential risk for legaltech startups. Your terms of service need to be airtight (and yes, have a real lawyer write them).
• The “good enough” trap. ChatGPT’s legal outputs are often directionally correct—maybe 80% of the time. That’s seductive. It’s also dangerous, because the 20% where it’s wrong often involves exactly the kind of nuanced, high-stakes situations where being wrong is catastrophic. Founders building AI legal tools need to resist the temptation to ship a product that’s “usually right.” In law, usually right is sometimes ruinous.
• Trust erosion. Every time a story like this goes viral, it erodes public and enterprise trust in AI tools broadly. If you’re selling an AI-powered product to legal departments, your sales cycle just got longer. Budget for that.

The Bottom Line

The Krafton case isn’t really about a CEO being reckless with a chatbot—though he clearly was. It’s about the dangerous gap between AI’s confidence and its competence. Large language models are extraordinarily good at sounding right. They are not good at being right in contexts that require deep domain expertise, situational judgment, and an understanding of consequences. For founders, the lesson is twofold: first, never let an AI tool be the final decision-maker on anything that could blow up your company. Use it for speed and exploration, then route the output through humans who actually know the domain. Second, if you’re building in the AI-assisted professional services space, the companies that win will be the ones that are brutally honest about what their technology can’t do.

The $250 million question—literally—isn’t whether AI belongs in legal and business strategy. It obviously does. The question is whether you’re building (and using) these tools with the discipline to know where the human handoff needs to happen. Chang Byung-gyu didn’t. Don’t be Chang Byung-gyu.