AI Board Minutes and Corporate Records: Discovery, Privilege, and Governance Risks

Corporate boards have embraced AI-powered meeting tools at a striking pace. Products that automatically record, transcribe, and summarize board and committee meetings are now standard features of many enterprise collaboration platforms. The pitch is compelling: directors get accurate summaries instead of contested recollections, general counsel can review preliminary minutes within hours instead of days, and the company builds a richer institutional record. What most companies have not fully absorbed is that this richer record cuts both ways. The same AI tools that capture deliberation with unprecedented fidelity may be generating documents that become exhibit A in the company’s next lawsuit.

This post examines the legal risks that AI-generated board minutes and transcripts create, what the Fortis Advisors v. Krafton decision tells us about how courts are treating AI records in corporate litigation, the uncertainty surrounding privilege and work product protection for AI-assisted materials, and the governance guardrails every board should put in place today.

What AI Meeting Tools Actually Produce

Before addressing privilege and discovery risk, it helps to be precise about what these tools generate. When a company deploys an AI meeting assistant during a board or committee meeting, it typically produces several layers of records:

• Audio and video recordings of the live session
• Verbatim transcripts of every spoken word, including off-the-cuff remarks, board member questions, dissenting comments, and management presentations
• AI-generated summaries that highlight key topics, action items, and decisions
• Draft minutes that the AI creates from the transcript before any attorney or corporate secretary reviews them
• Revision histories of the minutes as humans edit the AI draft

Each of these items is a document. Each can potentially be subpoenaed or obtained through discovery in litigation. Under the Federal Rules of Civil Procedure, discoverable material includes any non-privileged, relevant information in a party’s custody or control. There is no rule that says AI-generated drafts of board minutes are different from handwritten notes or a secretary’s draft minutes. If they exist and are relevant, they are fair game.

Traditional board minutes were drafted by a corporate secretary or counsel, reviewed carefully for legal accuracy, and maintained as the sole formal record of board deliberations. They were intentionally concise and disclosed decisions rather than the full texture of debate. AI-generated records do the opposite: they capture everything, including commentary that directors would never have approved for inclusion in formal minutes.

Fortis Advisors v. Krafton: AI Records as Litigation Evidence

The March 2026 Delaware Court of Chancery decision in Fortis Advisors LLC v. Krafton Inc. provided a stark demonstration of what AI-captured communications look like in court. The case arose from Krafton’s acquisition of the video game studio Unknown Worlds, where Fortis Advisors alleged that Krafton’s executives acted in bad faith to orchestrate the company’s operations in ways designed to deprive Krafton’s shareholders of a $250 million earnout payment.

What made the case notable beyond its facts was the evidence. The court relied on extensive ChatGPT chat logs generated by Krafton’s CEO. Those logs showed the CEO consulting a public AI chatbot at length about strategy — discussions that captured his intent and planning with a candor that would never have appeared in a traditional email chain or board memo. The AI conversations became an exhibit because they were not privileged: the CEO had used a public AI platform independently, not under counsel’s direction, and the platform’s privacy terms made clear that user inputs could be shared with third parties. There was no reasonable expectation of confidentiality.

The Fortis Advisors court did not create new law about AI board minutes specifically, but the case carried a lesson that every corporate secretary and governance counsel should internalize. AI-generated communications are documents. When they exist, they are discoverable. When they reflect candid internal deliberation, they can be devastating in litigation. And unlike a phone call that leaves no trace, AI meeting records are stored, dated, and searchable.

The Privilege Problem: Why AI Meeting Records Are Particularly Exposed

Attorney-client privilege protects confidential communications between a lawyer and client made for the purpose of obtaining or providing legal advice. Work product doctrine protects materials prepared in anticipation of litigation by or at the direction of counsel. Both privileges are well-established, but neither maps cleanly onto AI-generated board meeting records.

The attorney-client privilege gap. Most board meetings are not legal consultations. They are business meetings where management presents strategy, directors deliberate, and votes are taken. The presence of general counsel in the room does not automatically make everything said at the meeting privileged. Courts applying the dominant-purpose test ask whether the primary purpose of the communication was to obtain or provide legal advice. Board-level deliberations about strategy, financials, executive compensation, or M&A targets are business communications, even if a lawyer attends. The AI transcript captures all of it, privileged and non-privileged alike.

Even where individual portions of a board meeting may involve privileged legal advice from counsel — say, a discussion of pending litigation — the granular AI transcript can blur the line. Rather than a summary paragraph in traditional minutes that says “the board received a report from counsel regarding pending litigation,” the AI transcript records the actual exchange, including directors’ questions about litigation exposure and candid comments about liability risk. Extracting privileged portions from a 90-minute verbatim transcript is a much harder privilege log exercise than asserting privilege over a discrete set of counsel’s memoranda.

The work product problem. Work product doctrine protects material prepared in anticipation of litigation. AI-generated meeting notes are prepared in real time as a business record, not in anticipation of litigation. Even if litigation later arises from the subject matter discussed at the meeting, the routine creation of AI transcripts as a matter of course is unlikely to qualify for work product protection. The doctrine protects the mental impressions of counsel in preparing for litigation, not contemporaneous business records that happen to touch on topics that later become contested.

The third-party disclosure problem. A fundamental requirement of attorney-client privilege is confidentiality. Communications lose privilege when they are voluntarily disclosed to parties outside the attorney-client relationship. Many AI meeting platforms store transcripts and summaries in vendor cloud environments where the terms of service permit the vendor to use content for model training or to share data with affiliated entities. As the Heppner court analyzed in the privilege context (discussed more fully in a companion post), the privacy policy terms of public AI platforms may independently destroy the reasonable expectation of confidentiality that privilege requires. Even enterprise AI tools present this risk if the vendor contract does not include explicit zero-retention commitments and prohibitions on using client data for training.

The Candor Problem: When Good Notes Are Bad Evidence

One of the most frequently underappreciated governance risks of AI meeting records is the candor problem. Traditional board minutes were drafted to serve as the official corporate record, not as a verbatim account of deliberation. Experienced corporate secretaries and counsel understood that minutes should reflect decisions and significant discussion points — not every question, skeptical comment, or devil’s advocate argument that directors raised before reaching consensus.

The legal rationale for this practice is not to hide information. It is to preserve directors’ ability to engage in robust deliberation without transforming every question into an admission. A director who asks “has management stress-tested whether this acquisition price is defensible if the market turns?” is doing her job. If that question appears verbatim in a transcript that later surfaces in merger litigation, it may look to plaintiffs’ counsel like evidence that the board harbored doubts about fair value.

AI-generated transcripts do not apply this judgment. They capture everything. Humor, sarcasm, hypothetical concerns, and incomplete sentences all land in the record with equal weight to actual decisions. The challenge is compounded because the AI may summarize incorrectly or flatten nuance — characterizing a director’s probing question as a statement of concern, or attributing a comment to the wrong person. Unlike notes that were never taken, an erroneous AI summary is a document that must be produced and explained.

Regulatory and Fiduciary Dimensions

For regulated companies — banks, broker-dealers, investment advisers, and public companies — the use of AI meeting tools may create additional regulatory exposure. Securities regulations generally require retention of records of board-level communications and compliance-related deliberations. If an AI tool creates a more complete record of those deliberations than the company intended to maintain, the regulatory retention obligation may attach to the fuller record.

For public companies, the board’s deliberative process in M&A transactions, executive compensation decisions, and related-party transactions is frequently scrutinized in litigation. Delaware courts in particular have developed a rich body of case law examining whether boards satisfied their duties of care and loyalty, and those cases often turn on the quality and candor of boardroom deliberation as reflected in the record. An AI transcript that captures a perfunctory 20-minute discussion of a major transaction will be far more damaging in a Revlon or Caremark context than traditional minutes that documented the board’s process at a higher level of generality.

Governance Guardrails Every Board Should Implement

The answer to these risks is not to ban AI from the boardroom. AI meeting tools offer real value, and boards that refuse all technology will find themselves at a disadvantage in administrative efficiency without eliminating legal risk. The answer is thoughtful governance — rules that capture the benefits of AI while managing the privilege, discovery, and candor risks.

1. Adopt an AI meeting tools policy before deployment. The board’s governance committee, working with general counsel, should establish a written policy that governs which AI meeting tools may be used for board and committee meetings, what records those tools may generate and retain, and who controls each record type. A policy after a problem arises is worth far less than a policy before.

2. Classify records by type and retention period. Not all AI outputs are created equal. The policy should distinguish among audio recordings, verbatim transcripts, AI-generated summaries, and official minutes. For most companies, the right answer is that only the reviewed and approved official minutes constitute the corporate record. Audio recordings and verbatim transcripts should be deleted under a defined schedule after minutes are approved, unless litigation hold obligations require preservation.

3. Apply litigation holds to AI records. A litigation hold that covers emails and hard-copy documents but fails to address AI-generated meeting records is an incomplete litigation hold. When a hold is triggered — whether by a demand letter, the filing of a complaint, or a reasonable anticipation of litigation — counsel must identify and preserve all relevant AI meeting records, including verbatim transcripts and draft minutes, just as they would any other potentially relevant document.

4. Require attorney review before AI drafts become the record. AI-generated draft minutes should never become the official record without human review. General counsel or the corporate secretary should review every AI draft for accuracy, appropriate scope (capturing decisions without unnecessary candor about deliberative process), and potential privilege issues. The reviewed and approved minutes are the record. The AI draft is a working tool, not the end product.

5. Use enterprise-grade tools with contractual confidentiality protections. If your company uses AI meeting tools, the vendor contract must include: (a) a prohibition on the vendor using your meeting content to train AI models; (b) a commitment that data is not shared with third parties except as legally required; (c) defined data retention periods and deletion obligations; and (d) security and encryption standards appropriate for sensitive boardroom communications. Consumer-grade tools that lack these protections put every board meeting on terms that may eliminate any reasonable expectation of confidentiality.

6. Establish a pre-meeting protocol. Directors and management should be briefed on what AI tools will be used in the meeting and what records will be generated. Consider designating specified portions of board meetings — particularly attorney presentations on pending litigation — as off-record or as occurring in a format specifically structured to preserve privilege. This is not different in kind from the existing practice of meeting in executive session; it is an adaptation of that practice to the AI recording environment.

7. Train directors. Many directors do not know that their informal remarks during a board meeting may be transcribed verbatim and stored in a vendor’s cloud. Directors who understand the recording environment will engage more deliberately. That is not a restriction on candor — it is an exercise of the judgment that governance requires.

The Broader Lesson from AI Meeting Records

The Fortis Advisors case is one example of a broader pattern. Courts are receiving AI-generated records as evidence, and those records are often more revealing than the company intended. The risk is not confined to chatbot conversations like the one at issue in Fortis Advisors. Any organizational use of AI tools that generates records of deliberation creates the same exposure: the AI captures more, retains it longer, and stores it in formats that are easily searchable and producible.

Boards are not exempt from this dynamic. In many ways, they are among the highest-risk environments because the subjects discussed at board meetings — strategic decisions, litigation exposure, M&A activity, executive performance — are precisely the subjects most likely to become contested in future litigation.

The governance task is not to turn back the clock. AI meeting tools are here, they are improving, and they offer real advantages. The governance task is to ensure that the boards and companies using these tools have made deliberate choices about what records to create, how long to retain them, and how to structure the attorney-review process that stands between the AI’s raw output and the official corporate record. That deliberate choice — not the technology itself — is the exercise of governance.

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.