AI in M&A Due Diligence: When AI-Generated Deal Communications Become Litigation Evidence

Deal teams have embraced AI quickly. Associates use large language models to summarize due diligence findings. CFOs query ChatGPT about financial model assumptions. Business development executives ask AI assistants to draft deal rationales, assess earnout risk, and model post-closing integration scenarios. The efficiency gains are real and the tools are genuinely useful.

What is not yet widely understood — and what is increasingly documented in court decisions — is that those AI interactions may be discoverable in post-closing disputes. Chat logs with AI tools, documents generated through public AI platforms, and AI-assisted analysis produced during the deal process are creating a new category of litigation evidence that deal teams did not previously have to manage.

The Delaware Court of Chancery’s March 2026 decision in Fortis Advisors, LLC v. Krafton, Inc. made this concrete in a way that no amount of abstract warnings could. The court quoted at length from a CEO’s ChatGPT conversations. Those conversations became evidence in a $250 million earnout dispute, and they were central to the court’s finding of bad faith. The lesson for any company involved in M&A — as buyer, seller, or financial advisor — is that the AI tools you use during a deal and after it closes are generating a documentary record that did not exist five years ago.

This post examines the legal framework, the Krafton case in detail, the privilege issues, and what AI use policies for deal teams should look like.


The Fortis Advisors v. Krafton Case: What Happened

In 2021, Krafton, a South Korean video game conglomerate, acquired Unknown Worlds Entertainment — the studio behind the hit game Subnautica — for $500 million upfront with a potential additional $250 million earnout contingent on the studio meeting specified performance targets over a multi-year measurement period. As part of the deal, Krafton contractually guaranteed that the studio’s founders and CEO would retain operational control of the business and could only be removed for contractually defined “cause.”

The relationship broke down. Krafton terminated the studio’s key executives and asserted operational control over the studio in ways the sellers alleged violated the agreement. Fortis Advisors, the seller representative, sued in the Delaware Court of Chancery for breach of the Equity Purchase Agreement.

During discovery, Krafton’s CEO’s ChatGPT conversation logs were obtained and became part of the evidentiary record. According to the court’s March 2026 opinion — widely reported by Harvard Law School’s Corporate Governance Blog and multiple law firm analyses — the CEO had turned to ChatGPT to develop strategy around the earnout situation. ChatGPT reportedly advised that the earnout would be “difficult to cancel” but nevertheless provided the CEO with a strategy for attempting to do so.

The court quoted those AI-generated exchanges extensively. Because the CEO had obtained this strategic guidance directly from a public AI tool rather than from legal counsel, the attorney-client privilege did not apply. The communications were fully discoverable, and the AI’s responses — and the questions the CEO had asked — provided direct evidence of the buyer’s intent to circumvent the earnout.

The court found that Krafton had breached the Equity Purchase Agreement, ordered the reinstatement of the seller’s CEO with full operational authority, enjoined Krafton from using the studio’s board to circumvent that authority, restored the studio’s access to its publishing platform, and — critically — equitably extended the earnout measurement period by 258 days to account for the time during which Krafton’s improper conduct had impaired the studio’s ability to hit its targets.

The lesson embedded in this outcome is not subtle: when a senior executive uses a public AI tool to strategize about a contractual dispute, those conversations may become the most damaging documents in the litigation.


Why AI Conversations Are Discoverable

Federal and state civil discovery rules require parties to litigation to produce documents and communications that are relevant to the claims or defenses at issue. The default rule is that relevant, non-privileged documents are discoverable. The burden is on the party asserting a privilege or protection to establish its applicability.

The Attorney-Client Privilege Does Not Apply

Attorney-client privilege protects confidential communications between a client and an attorney made for the purpose of seeking or providing legal advice. AI tools are not attorneys. Queries to ChatGPT, Claude, Gemini, Copilot, or any other public AI platform do not involve an attorney and are not made in the context of an attorney-client relationship.

A February 2026 federal court decision in the Southern District of New York — United States v. Heppner — addressed this directly. The defendant had used Anthropic’s Claude AI platform after receiving a grand jury subpoena to generate 31 documents analyzing his legal exposure, defense theories, and strategic options. He subsequently attempted to claim privilege over those documents.

The court held that the documents were not protected by either the attorney-client privilege or the attorney work product doctrine. The AI-generated documents did not involve legal advice from an attorney. The work product doctrine did not apply because the documents were not prepared by an attorney or at an attorney’s direction. The fact that the defendant later sent some of the documents to his lawyer did not retroactively create privilege — privilege must attach at the time of the communication, not after the fact.

The practical implication for deal teams: if your CFO uses ChatGPT to analyze potential liability under an indemnification clause, that analysis is not privileged. If your business development team uses an AI tool to model earnout scenarios, those outputs are not privileged. If your integration team queries an AI assistant about post-closing covenants and their enforceability, those conversations are not privileged.

The Work Product Doctrine Does Not Apply

The work product doctrine protects materials prepared by an attorney (or at an attorney’s direction) in anticipation of litigation. Deal communications and analyses generated by businesspeople using public AI tools — before litigation is anticipated, outside of attorney direction, and for ordinary business planning purposes — do not meet this standard.

Even where litigation is anticipated, work product protection for AI-generated materials requires that the materials were prepared at the direction of counsel, not independently by executives using public tools. The Heppner decision specifically rejected the argument that independently prepared AI-assisted analysis could be shielded as work product.

The Third-Party Waiver Problem

There is an additional concern specific to public AI platforms. When a user submits a query to a public AI tool, the query and response may be stored, retained, and potentially accessible to the platform provider. Under traditional privilege doctrine, disclosing privileged communications to a third party — including a service provider — waives the privilege. The extent to which disclosure to public AI platforms constitutes a waiver of any otherwise-applicable privilege is an active area of legal analysis, but the general principle counsels against using public AI tools for genuinely sensitive legal analysis.

Enterprise AI tools with appropriate data protection agreements and confidentiality provisions present a different analysis. But the default-access public versions of these tools do not.


What AI-Generated Deal Communications Look Like as Evidence

The Krafton case demonstrates one failure mode — executive AI conversations that reveal bad-faith intent. But the broader discoverable record created by AI use in deal processes extends considerably further.

AI-assisted due diligence summaries. Law firms and financial advisors increasingly use AI tools to summarize and analyze due diligence data rooms. Those AI-generated summaries — what the tool flagged as risks, what it characterized as acceptable, what it missed — are potentially discoverable in post-closing disputes about what was known or disclosed.

Chatbot logs from negotiation-related queries. Deal team members who query AI tools about negotiation strategy, contract interpretation, market standards for deal terms, or post-closing obligations are generating logs that may be relevant to questions about what the parties intended or understood.

AI-assisted valuation and financial model commentary. AI-generated commentary on financial models, assumptions, or valuation methodologies may be relevant in post-closing disputes about purchase price adjustments, earnout calculations, or representations about the target’s financial condition.

AI drafting interactions. When deal documents are drafted with AI assistance, the drafting process itself — including what the AI was asked to draft, what alternatives it generated, and how those were modified — may be relevant to questions of contract interpretation. Ambiguous deal terms can be illuminated or complicated by evidence of how those terms were generated.

Integration planning AI outputs. Post-closing integration planning that uses AI tools to analyze workforce decisions, technology consolidation, or operational changes may generate documents relevant to post-closing covenant compliance claims, WARN Act disputes, or other post-closing litigation.


The Emerging Law on AI and Privilege

Courts are just beginning to develop doctrine on how privilege and work product apply in AI-assisted legal work. The early decisions suggest several principles that deal teams should understand.

Publicly accessible AI platforms generate non-privileged records. This is the core holding of Heppner and the implicit lesson of Krafton: using public AI tools for legal analysis, strategic planning, or contract interpretation does not create privileged communications.

AI outputs sent to counsel do not become privileged after the fact. Transmitting an AI-generated document to your attorney does not make the document privileged. Privilege attaches to the attorney-client communication, not to underlying documents that preceded it.

Attorney-supervised AI use may be different. The analysis changes when AI tools are used under attorney direction as part of a legal engagement. An attorney who uses AI to assist in drafting a legal memo or analyzing documents as part of a privileged representation can make a stronger privilege argument. The distinction is between AI used by businesspeople for business purposes and AI used by or under the direction of counsel for legal advice purposes.

The law is actively developing. In early 2026, proposed amendments to the Federal Rules of Civil Procedure addressing AI-related discovery were circulating for comment, and at least two federal circuits had cases involving AI-generated document privilege working through the appellate process. The doctrine will develop quickly over the next several years, and today’s analysis represents the current state, not a stable endpoint.


AI Use Policies for Deal Teams: A Practical Framework

Every company engaged in active M&A activity — whether as a frequent acquirer, a company evaluating a sale, or a portfolio company managing an active deal process — needs an AI use policy specific to the deal context. Here is a practical framework.

1. Classify AI Tools by Risk Level

Not all AI tools present the same risk. Enterprise AI platforms deployed under a written agreement with appropriate data protection provisions, where the company retains ownership of its data and the platform does not use inputs for model training, are materially lower-risk than public consumer AI platforms. Deploy enterprise tools for sensitive deal work; prohibit or restrict public AI tool use for any deal-sensitive analysis.

The bright line for privilege purposes is attorney involvement. When deal team members need analysis of legal exposure, contract interpretation, indemnification risk, or regulatory issues, that analysis should flow through or be directed by counsel. An attorney who uses AI to assist in producing a privileged legal analysis, under client instruction and within the attorney-client relationship, creates a materially more defensible privilege position than an executive using the same tool independently.

Practically: establish a protocol that routes legal questions in the deal context to counsel, not to AI tools. The efficiency gained by skipping counsel for legal questions is likely to be a poor tradeoff if those AI interactions become exhibit A in post-closing litigation.

3. Implement Document Retention Policies That Cover AI

Most deal teams have document retention policies covering email, internal messaging platforms, and paper documents. Far fewer have extended those policies to AI interaction logs. As AI tools become part of standard deal workflows, retention and litigation hold policies need to explicitly address:

  • Chat logs and conversation histories in AI tools
  • AI-generated drafts and outputs saved to deal team workstreams
  • Prompts submitted to AI tools as part of due diligence analysis
  • AI-generated summaries of due diligence findings

When a litigation hold is triggered — including litigation holds triggered by the possibility of post-closing disputes that arises in virtually every significant M&A transaction — the hold must cover AI-generated materials.

4. Treat AI Interactions as Business Records During Due Diligence

AI-assisted due diligence creates documentation of what was reviewed, what was flagged, and what conclusions were reached. This documentation is operationally valuable and potentially legally significant. A seller who uses AI to analyze a data room and the AI generates a summary that says a specific risk was “acceptable” or “minimal” has created a document that may be discoverable and may speak to the seller’s own assessment of the risk it was disclosing (or not disclosing).

Deal teams should approach AI-generated due diligence summaries with the same care they apply to written analyst memos: as documents that will exist in perpetuity and may be reviewed in the context of post-closing disputes.

5. Restrict Executive Use of Public AI Tools for Deal Strategy

The Krafton decision is a cautionary tale specifically about executive-level use of public AI tools to strategize about contractual obligations. A written policy restricting the use of public AI platforms for any analysis related to deal strategy, earnout structures, post-closing obligations, or potential disputes is a straightforward mitigation step. This is not about restricting AI use generally — it is about ensuring that sensitive strategic analysis does not occur in a venue that creates an unprotected documentary record.

6. Address AI Use in Deal Documents

Term sheets and definitive acquisition agreements should be updated to address AI-related issues:

  • Representations and warranties about AI system use in the target’s business (covered more fully in Posts 110 and 111 in this series)
  • Retention and production obligations with respect to AI-generated communications in the event of post-closing disputes
  • Definitions of “written communications” for purposes of merger agreement provisions that turn on written consent or approval

These are not yet standard provisions in most deal documents, but the legal landscape is developing quickly enough that sophisticated deal teams should be raising them.

7. Train Deal Teams

Privilege education in the deal context has historically focused on email — who to copy, when to copy counsel, how to mark communications. That training needs to be updated to address AI tools. Deal team members should understand:

  • That public AI tool interactions are not privileged
  • That AI-generated documents can be discovered in litigation
  • The company’s protocols for routing legal questions through counsel
  • The document retention obligations that apply to AI-generated materials

M&A Counsel Should Affirmatively Inquire About AI Use

For the lawyers advising on M&A transactions: the Krafton case and related decisions create an affirmative obligation to inquire about AI tool use by deal team members during the transaction.

This means asking at the outset of an engagement: what AI tools is the deal team using? How are AI-generated materials being handled and retained? Are there existing interactions with public AI platforms related to this transaction that may be relevant to our privilege analysis?

In post-closing disputes, early investigation of AI tool use history should be part of the initial case assessment. The evidentiary record now extends beyond email inboxes and document management systems to include AI tool logs — and in some cases, those logs will be among the most significant evidence available.


The Bottom Line

AI tools have become part of the deal process. That is not going to reverse. What needs to change is how companies manage the documentary record those tools create. The distinction between enterprise AI deployed under appropriate data protection terms and used within attorney-supervised workflows versus public AI tools used independently by businesspeople for legal and strategic analysis is the most important line deal teams need to understand.

The Krafton decision demonstrated with considerable clarity what happens when that line is crossed: a CEO’s ChatGPT conversations become a judicial opinion. For any company with meaningful M&A activity — as acquirer, target, or post-closing party to earnout or indemnification disputes — building an AI use policy for deal processes is not a discretionary item. It is a baseline risk management requirement.


This post is for general informational purposes only and does not constitute legal advice. Reading this post does not create an attorney-client relationship. If you have questions about your specific situation, consult a qualified attorney.



Leave a Reply