Directors, AI and the board room: what the courts are now telling boards

If directors are using AI to support them in the boardroom or to work out "what to do" about a difficult issue, now is the time to put structures and protocols around the use of AI in this context. This article considers two judgments delivered in 2026 which discuss and highlight the potential benefits and pitfalls of using AI in the boardroom.

Directors’ duties apply to AI-assisted decisions

Directors' duties under the Corporations Act 2001 (Cth) (Corporations Act) and at common law are not new. Directors must act with care and diligence, in good faith in the best interests of the company and for a proper purpose.

However, the prevalence of AI and the rate at which AI is being both used and developed overlays a new lens through which directors must consider how they discharge their duties. This is particularly relevant in the context of the core directors' duty under section 180 of the Corporations Act, to exercise their powers and discharge their duties with a degree of care and diligence that a reasonable person would exercise if they were a director of a company in the company's circumstances and had the same responsibilities within the company. As discussed in our recent Clayton Utz Insights article, it is clear that boards must understand how AI is used in their organisation, and oversight cannot be delegated away.

There are now a number of cases which comment on, or consider, the use of AI in the boardroom. Earlier this year, in Australian Securities and Investments Commission v Bekier (Liability Judgment) [2026] FCA 196 (Star Entertainment), the Federal Court indicated that, with appropriate controls and oversight, AI could be a useful tool to support directors. However, a recent decision of the Supreme Court of New South Wales (NSWSC), WLLHLL Holdings Pty Ltd v Drebber (No 2) [2026] NSWSC 800, shows the risks of directors using AI in the governance of their company.

When AI helps: ASIC v Bekier (Star Entertainment)

In Star Entertainment, the Federal Court took a constructive view of the potential benefits of AI (while still cautioning against potential risks). As has been widely reported in the context of that decision, the Federal Court rejected the argument that the volume of material excused directors from reading it, and repeated that a board can, and should, take steps to control the volume of information it receives.

Importantly, the Federal Court accepted that directors now commonly use AI to navigate board materials and that there is nothing inherently objectionable in obtaining such assistance. However, the Federal Court stated that what ought not occur is that AI becomes an excuse for poor information discipline, or a "quiet normalisation of private reliance ... upon computer-generated distillations, unregulated by any agreed policy". As the Federal Court put it, "the use of technology may assist comprehension, but it cannot displace judgment".

When AI goes wrong: WLLHLL Holdings v Drebber

The recent NSWSC case related to a shareholder oppression claim in respect of Lanmar Pty Ltd (Lanmar), which was owned equally by three parties, each of whom is controlled by a director. While ultimately a shareholder oppression case, there is an instructive discussion on the potential pitfalls of relying on AI.

The case concerned a dispute between the three director-shareholders about their contributions to, and entitlements owing from, Lanmar. The result of the dispute was that two of the director-shareholders decided to remove the third director-shareholder.

Rather than seeking legal advice about removing the director-shareholder, they asked ChatGPT.

Importantly, ChatGPT incorrectly treated the fellow director-shareholder as an underperforming employee and wrongly suggested a course of action by reference to employment law principles rather than focusing on principles of oppression under Australian corporate law. The NSWSC commented that the step of "forcing/commanding ... plainly did not reflect the relationship between directors or shareholders under Australian corporate law".

Justice Black observed that ChatGPT "did not prove to be a prudent choice of adviser" on Australian corporate law. By framing the issue as performance management and employee termination, the AI output distracted the directors from the real risk of oppressive conduct under section 232 of the Corporations Act.

The findings noted that the ChatGPT output even suggested diverting business to other companies, a step his Honour said could have resulted in a breach by the directors of their duties, and cited a clause that did not exist in Lanmar's constitution.

Ultimately, his Honour held that this constituted conduct of the affairs of the company (by the majority shareholders) in a manner that was unfairly prejudicial, or oppressive, to the other director in their capacity as a shareholder.

The contagion risk for fast-growing companies

The Lanmar scenario is a familiar one. AI is increasingly used across companies. However, this case highlights the contagion risk associated with relying on AI output without verification as a foundation for a course of action. In the Lanmar case, the directors based their course of action on flawed AI-generated advice. While each later step may have looked reasonable in isolation, the course of action was ultimately found to be oppressive and to have resulted in a potential breach of duty.

AI output can also be confidently wrong or quietly incomplete. An output may appear complete yet omit key material without any sign of the gap. Directors who rely on it without checking may find it difficult to establish that they acted with care and diligence in the event that they face a claim for a breach of directors' duties.

The practical answer is simple – directors must not follow an AI-generated decision or output without forming their own view and assessment. AI can be utilised to assist directors to discharge their duties, but it is not a substitute for professional advice and independent board oversight and decision-making.

Future-proofing decisions: keep the decision trail

If AI-drafted documents end up in the board’s records, they may show what was decided but not why, or who exercised judgment. That matters if decisions are later challenged. In Star Entertainment, the Federal Court relied heavily on minutes and emails, which disclosed "little by way of sustained scrutiny". In Lanmar, the AI-assisted plan was exchanged through private email accounts to keep it "off the books", yet was produced in evidence and shaped the Court's findings.

Boards should ensure minutes record the material considered, the questions asked and the reasons for the decision. Where AI was used, record how and for what, and keep the prompts and outputs that informed a material decision.

In practice, a record of the decision making process should allow a reader to reconstruct how AI informed the outcome. For any material decision, that means:

• retaining the prompts used, the outputs the board actually relied on, and a note of the independent review undertaken, including who checked the output, against what sources, and whether professional advice was obtained;

• minutes should clearly distinguish AI-generated material from the directors' own reasoning. Rather than adopting AI-drafted analysis wholesale, they should record where directors tested, questioned or departed from the output and why. This shows that the board exercised its own judgment rather than simply endorsing a machine-generated conclusion; and

• boards should be alert to informal channels within the organisation. As Lanmar illustrates, circulating AI-assisted analysis through private email accounts does not keep it out of evidence. It simply removes it from the board's control and makes the decision harder to defend.

What boards should do now

In Star Entertainment, the Federal Court commented that the use of AI should be "controlled and transparent" and that it must be governed by "formal adoption of policies, rather than just wink at informal 'shadow' use".

Technology (and its application) continues to evolve rapidly. To support directors in demonstrating they have discharged their duty to act with due care and diligence, boards should:

• Understand when and how AI is being used not only by the board but also across the business.

• Adopt a formal AI-use policy for directors, management and employees, rather than tolerate informal "shadow" use. Given the fast evolution of these tools, an AI policy cannot be a "set and forget" policy. The use of AI across a company should be tested continually, including to identify potential risks and weaknesses in the tools being deployed.

• Treat AI summaries as a navigation aid, not a substitute for reading key papers.

• Apply the four requirements identified in Rogers v McDonald’s Australia Ltd (AI-use) [2026] FCA 1264 to any AI tool utilised, as summarised in our recent Clayton Utz Insights article:

  • Traceability: it must preserve traceability to the primary records;

  • Testability: its outputs must be capable of being tested and corrected;

  • Confidentiality: it must protect privileged and confidential material where necessary; and

  • Human supervision: it must remain subject to appropriate human supervision.

• Confirm that confidential company information (such as board papers and commercially sensitive strategic information) is entered only into approved AI tools (and not public AI tools such as ChatGPT).

Act now

Boards that set clear, transparent rules for AI use, and keep a genuine record of their own reasoning, will be best placed to demonstrate they have appropriately discharged their directors' duties.