When ChatGPT writes the strategy, directors carry the judgment

The most important lesson from the New South Wales decision in Lanmar Pty Ltd (No 2) is the court's focus on process. Two directors of Lanmar Pty Ltd relied on ChatGPT to develop a plan for removing a fellow director and shareholder. They followed the resulting “HR problem resolution” plan. The court found that their conduct was oppressive and that they had breached their duties as directors. It appointed a receiver to arrange the sale of the company's shares on the open market.

It would be easy to classify this as another case of AI inaccuracy. The more important lesson, however, concerns reliance. The criticism was not simply that the model provided incorrect or incomplete legal advice. The directors had framed the question incorrectly and then acted on the answer without inserting independent human judgment between the AI output and the decision. The case illustrates why directors must exercise independent judgment when using AI: responsibility for a decision remains with the people who make it, not with the tool that helped prepare it.

The liability lies in the reliance

The case involved a small company with three equal economic interests, held through separate entities controlled by the individuals involved. That structure will be familiar to many mid-sized businesses: each principal is closely involved in the operation, external legal advice is often cost-sensitive, and a governance issue can easily be framed as a performance problem. The majority directors approached their fellow director and shareholder as though he were an underperforming employee. The court treated the dispute as a matter of corporate governance and oppressive conduct under Australian law. ChatGPT, by contrast, approached it as an employment law problem.

For a European or Middle Eastern organisation, the precise legal implications will differ. Directors' duties, shareholder remedies and the legal effect of court judgments vary by jurisdiction. The Australian decision should therefore be understood as a governance signal rather than a precedent that automatically applies to every board. What travels across jurisdictions is the process the court examined. The issue was not simply whether the model's answer was accurate. It was whether anyone had recognised the nature of the decision, checked the advice against the applicable legal framework, and taken responsibility for the step between advice and action.

This is where AI use moves from a productivity question to a governance question. When AI-assisted material reaches the board without a record of who reviewed it, what they challenged, and what they changed, the organisation creates a governance risk. The decision may appear adequately supported at the time, but the organisation could struggle to explain later how it was reached and who took responsibility for it.

Boardrooms are already using these tools quietly

In a company employing a few hundred to a few thousand people, the pattern seen in Lanmar may be closer than the different legal jurisdiction suggests. Many businesses do not have in-house counsel. Someone preparing a board paper on a dispute, a restructuring or a supplier claim may quietly use an AI model to structure the argument. They may not tell the board because no one has asked. The question may only arise after something has gone wrong.

The exposure extends beyond the boardroom. An operations director who relies on a model to frame a response to a supplier or a proposed customer settlement is engaging in the same kind of reliance, often with even less scrutiny. The risk emerges wherever a model helps frame a decision with legal or regulatory consequences.

Most acceptable-use policies answer whether employees may use AI. They do not necessarily specify what must be reviewed before AI-assisted work becomes the basis for a board decision, who is accountable for that review, or where it must be documented. This creates an accountability gap. Our advisory services focused on board-level risk routinely examine where AI-assisted judgment is already taking place without clear ownership or review requirements.

What documented human review looks like

Documented review may sound like compliance overhead. In a mid-sized business, however, it can be as simple as three sentences in the relevant board paper: what the AI was asked, what it produced, and what the accountable reviewer rejected or changed after checking the output. The length matters less than the fact that a named person takes responsibility.

For material decisions, that person should be someone whose role already carries responsibility for the outcome: the relevant director, the chief operating officer or the head of risk, rather than simply the person who generated the draft.

The reviewer should also ask a different question from the one the AI answered. In Lanmar, the tool framed the situation as though a fellow director and shareholder were an employee. A competent reviewer would have questioned whether the issue was primarily one of shareholder rights and corporate governance, rather than employment management. The value of review lies less in checking every citation than in recognising the category of decision and identifying the question the model missed.

For organisations seeking to integrate this into their broader compliance obligations, our framework services place the review process within existing board reporting and governance arrangements, rather than creating a separate AI policy that may be overlooked in practice.

A question for the next board pack

The Lanmar judgment might be read in some quarters as an Australian corporate law issue with limited relevance elsewhere. Its more durable lesson is straightforward. Boards are increasingly making decisions based on material prepared with AI that can sound confident while missing the question the board actually needs to answer.

The next time a board pack contains a restructuring plan, a response to a dispute or a summary of legal exposure, one question deserves a place near the front: Was any part of this material prepared or assisted by AI, and who has confirmed that they reviewed it?

That question costs little to ask, yet it can change how decisions are prepared. Organisations that cannot answer it may eventually be asked to explain their processes by a court, a regulator or an insurer. At that point, saying that the decision was based on an AI-generated recommendation will not be enough.