AI in the Boardroom: 
A Governance Framework for harnessing the power of AI

Transcribing or taking notes of Board meetings

• Can assist the Company Secretary in preparing accurate minutes.
• May capture detail that handwritten notes miss.
• Directors may wish to have notes of the meeting (to be later deleted or otherwise dealt with in accordance with Company policy relating to notes of meetings).
• Increasing efficiency within Company Secretariat, freeing up time for more value-adding work.
• May produce a more consistent format and structure across meetings.

Discoverability

Could the recording and/or transcript be discoverable in the context of litigation or compulsorily produced to a regulator?

Short answer, yes. Every off-the-cuff remark, tentative view and half-formed question could be produced to a regulator or in a courtroom. This would be a material and unwelcome change from the current position where minutes are the primary record.

Accuracy

Whilst the technology is continuing to improve, current platforms often find it hard to distinguish between different people and find discussions (or people talking over each other) difficult to accurately record.

Chilling effect

Directors may self-censor if they know every word is being recorded, undermining the quality of debate and challenge. This has long been the argument against recording Board meetings (even without the use of AI).

Human judgement is essential

Preparing minutes of meeting is an art – AI-generated minutes may capture too much, too little or the wrong things, as could a human. Naturally, any minutes produced by AI would require review from the Company Secretary, who will understand nuance and broader context that AI does not.
The key risk is how AI is used to produce the minutes, recognising the significant risks with using AI to record the meeting. However, sensible alternatives exist to make this a palatable option.

Assisting Directors in reviewing or analysing Board papers

• AI tools could help summarise key issues, flag inconsistencies or highlight areas requiring attention across what are often hundreds of pages.
• AI can summarise what the Board has already been advised of on a particular topic.

It is worth noting that management could provide the above information within the Board paper in the first instance, either with or without the assistance of AI (but recognising that AI certainly provides efficiency gains – e.g. where several Board papers need to be summarised to remind Directors what has previously been reported on a topic).

Duty to read and understand

Directors cannot outsource their duty of care to an algorithm.
In fact, Justice Lee in the recent Star Entertainment proceedings took the opportunity to say that a director needs to read the material personally and not just rely on AI to flag what is of significance in board papers.
If a Director relies on an AI-generated summary and misses a critical issue in the full papers, they will be assumed to have read the full paper notwithstanding they might not have done so. This is the case whether the Director is reading an AI-generated summary or management-prepared summary.

Quality of and variation in AI output

Current AI tools can hallucinate, miss context or mischaracterise nuance. An AI-prepared executive summary is not the same as one prepared by management who understand the business context – all of which can be addressed with human oversight and checks. Is a director entitled to “rely” on the AI summary in the same way that he or she could rely on information provided by management? That question is currently unanswered, but quite possibly, no.

Relatedly, if each director is asking AI to generate a summary, there is a real risk that each director is reading and relying on a different summary – this may impact the effectiveness of debate, but also create issues if any dispute arises around what has occurred at a Board meeting and how the Board as a whole has approached critical decision-making.

False confidence

AI tools may give directors a sense of comfort that they have adequately engaged with the material, when in fact they have engaged with an algorithm’s interpretation of it. Ultimately, a key risk sitting across the use of AI in the Boardroom is that it subconsciously undermines human diligence.

Discoverability

As above, if Directors are asking AI to answer questions about Board papers, could those questions (and AI’s responses) be discoverable? Directors may believe they are sharing questions, reflections and worries with ‘AI’ but could those remarks ultimately be read one day by a regulator or plaintiff? Short answer, yes.

Preparing Board papers

• Could help management synthesise large volumes of data into Board-ready formats.
• May improve consistency and structure of papers across the organisation.

Accuracy and accountability
Management always remains accountable for the accuracy of AI-drafted papers and the executive officers expose themselves to risk where there is an overreliance on AI leading to misinformation or material omissions in the Board reporting. If the directors are on notice that the papers were substantially AI-generated without adequate human review, their reliability may be called into question, such that directors may not acting reasonably be entitled to rely on them.

A framework would need to be developed to support the use of AI in preparing Board papers, to protect Directors and other officers. This certainly seems to us to be lower risk that Directors generating their own summaries.