For years, much of the discussion about artificial intelligence and corporate governance has concentrated on what AI might help management or the company secretariat produce.
Can AI draft a board paper? Summarise a report? Prepare first-cut minutes? Identify actions? Analyse a large document? Improve an executive summary?
These are legitimate questions. Used carefully, AI may make many governance processes faster and more efficient.
But a recent Australian example suggests that we may be asking too narrow a question.
Perhaps one of AI’s more significant contributions to board effectiveness will not be helping management produce more information.
Perhaps it will be helping directors interrogate the information they already have.
That distinction deserves attention.
The board information problem
Modern directors rarely suffer from a shortage of information.
If anything, the opposite is true.
Boards routinely receive substantial packs containing financial results, management reports, risk information, strategy papers, committee material, project updates, regulatory developments and recommendations requiring decision.
The director’s challenge is not simply to read this material. It is to place it in context.
What did management tell us six months ago?
What assumptions supported the original decision?
What did the board resolve?
What conditions were attached to its approval?
Were particular actions requested?
Has management now addressed them?
Does the current recommendation differ from what was previously proposed?
Is an emerging trend visible across several previous reporting periods?
Those questions require directors to connect information across time and across documents.
And that is where an interesting application of AI is emerging.
Telstra’s Board AI agent
In June 2026, Telstra Group Company Secretary and Legal Executive Craig Emery described publicly how Telstra had introduced a purpose-built Board AI agent to assist directors preparing for board and committee meetings.
The catalyst was revealing.
A director was not asking for more information. The director wanted help finding the right information, quickly, from material the board already had.
Telstra’s response was to develop an AI tool allowing directors, using prompts and conversational interaction, to search and interrogate board information. The agent can, for example, summarise relevant previous board or committee papers and minutes, identify resolutions previously made and help determine whether actions requested by the board have been addressed in current material.
Importantly, Telstra has drawn explicit boundaries around the technology.
The agent does not make decisions. It does not provide advice. It does not replace the responsibilities of directors.
Its role is to support thinking–not undertake the thinking itself.
That distinction is fundamental.
A “walled garden”, not an open AI experiment
Board information is among the most sensitive information held by an organisation.
Telstra therefore deliberately constrained what its agent can access.
According to the company, the system operates within a “walled garden” containing material already provided to the board, including previous board and committee papers and minutes, actions and resolutions, group policies, annual reports and previously released ASX announcements.
It sits within Telstra’s protected internal environment and is not connected to public or external large language models. Telstra says the system has also been tested against risks such as hallucination and operates within its existing privacy, cybersecurity and data-governance framework.
That architecture is as important as the AI capability itself.
It illustrates an important governance principle: the question should not simply be “Can AI do this?”
It should also be:
What information can the system access, where does that information go, what controls apply, how can outputs be validated and who remains accountable?
From producing information to interrogating it
This is where the Telstra example becomes particularly interesting.
There is a danger that organisations respond to generative AI by simply producing things faster.
More reports.
More analysis.
More summaries.
More slides.
More board papers.
That may increase productivity without necessarily improving governance.
Indeed, if an organisation already suffers from overly long board packs or poorly focused reporting, using AI merely to generate information faster could magnify the problem.
The better opportunity may lie elsewhere.
AI can potentially help directors move more efficiently through an established body of trusted corporate information so they can spend more of their limited preparation time asking questions such as:
- What has changed?
- What does not reconcile?
- What assumptions have moved?
- What previous commitments remain outstanding?
- What alternative explanations should we consider?
- What information is missing?
- What should we challenge?
In other words, AI may have greater governance value when it helps create better questions, rather than merely more answers.
This changes the board-reporting conversation
There is also a direct connection between this development and the longstanding challenge of improving board papers.
A well-designed AI retrieval tool cannot fully compensate for poorly structured source material.
If board papers contain unclear recommendations, inconsistent terminology, vague actions, buried assumptions or no clear distinction between matters for decision and matters for noting, those weaknesses do not disappear because AI is introduced.
They may become easier to find – but they remain weaknesses.
The future of AI-assisted governance therefore reinforces rather than diminishes the importance of good board-reporting disciplines.
That includes:
- a clear purpose for each paper;
- concise executive summaries;
- explicit recommendations;
- identified assumptions and uncertainties;
- genuine alternatives where appropriate;
- clear treatment of material risks;
- traceable decisions and actions;
- consistent terminology and document structures; and
- reliable governance records over time.
Structured, disciplined board information becomes more valuable when directors – and eventually AI tools acting under carefully designed controls – can retrieve and connect it efficiently.
The better the underlying governance information architecture, the greater the potential value of AI layered over it.
A different role for the company secretary?
This development also has implications for company secretaries and governance professionals.
Traditionally, the company secretary has often served as an important source of institutional memory.
Who remembers why the board made that decision?
Where is the paper from two years ago?
Was that condition included in the resolution?
Did the committee previously consider this issue?
What follow-up did the board request?
Those questions will not disappear.
But AI may increasingly assist with the retrieval component.
That could allow the company secretary to spend less time locating information and more time applying the professional judgement that technology cannot replicate: understanding context, identifying governance implications, recognising sensitivity, advising on process and ensuring that the board receives information in a form that supports effective decision-making.
In that sense, AI need not diminish the company secretary’s role.
Used well, it may elevate it.
The board still has to think
There is an obvious caution.
An AI system capable of summarising large amounts of historical board material can create an impression of authority that its output may not deserve.
A concise answer can still be incomplete.
A confident answer can still be wrong.
Historical records themselves can contain ambiguities.
Context matters.
Legal professional privilege, confidentiality, privacy and cybersecurity remain significant considerations.
And directors cannot outsource their duties or judgement to an algorithm.
The AICD’s current guidance reflects these concerns. Its research indicates that AI use by Australian directors and boards remains at an early stage, with barriers including access to secure enterprise systems, internal policies and legal risks. It also identifies a “two-speed” dynamic in which individual or “shadow” AI use can move ahead of formal collective board adoption.
The AICD has separately warned against directors placing confidential board material into inappropriate public AI tools and stresses that AI-generated summaries or analyses should not substitute for directors’ own review and interrogation of board papers.
Similarly, updated joint guidance from the AICD and Governance Institute of Australia on AI and board minutes emphasises that potential efficiency benefits must sit alongside human oversight, professional judgement and appropriate safeguards.
These are not arguments against AI.
They are arguments for governing it properly.
Three principles worth retaining
The Telstra experience points towards three particularly useful principles for boards considering similar applications.
First, create a trusted environment.
Board-level AI should operate within a secure information architecture appropriate to the sensitivity of board material. Consumer AI tools and uncontrolled “shadow AI” use present very different risks from purpose-built or properly configured enterprise environments.
Second, make outputs capable of being tested.
Directors need to understand where information has come from and be able to verify it against source material. AI should strengthen informed challenge, not replace it with another black box.
Third, keep judgement and accountability human.
Technology can retrieve, summarise, compare and identify patterns. It may increasingly become very good at doing so.
But responsibility for the decision remains with the board.
That boundary matters legally, practically and culturally.
From “How do we govern AI?” to “How can AI help us govern?”
The first phase of board engagement with AI has understandably concentrated on oversight.
What is management doing with AI?
What risks arise?
Do we have an AI policy?
What controls and accountabilities are required?
Those questions remain essential.
But a second question is emerging:
How might AI improve the board’s own work?
The AICD has itself characterised the emerging conversation in similar terms–moving from “how do we govern AI?” towards “how does AI help us govern?”–while emphasising that Australian board practice remains relatively limited and evolving.
Telstra provides one practical answer.
Not an AI director.
Not an automated decision-maker.
Not a machine replacing board judgement.
Instead, a carefully bounded tool helping directors find, connect and interrogate trusted information more effectively.
That may sound less dramatic than some predictions about AI and corporate governance.
It may also prove considerably more useful.
The real opportunity
Boards already possess enormous amounts of information.
The problem is increasingly one of attention, retrieval, context and judgement.
The most valuable AI application in the boardroom may therefore not be one that generates another paper.
It may be one that helps a director ask:
“What do I need to understand before I make this decision – and what should I be challenging?”
If AI can help answer the first part without taking responsibility for the second, we may be looking at a genuinely useful evolution in board practice.
The objective should not be to automate governance.
It should be to create more space for the things that good governance has always required: informed judgement, constructive challenge, institutional memory, curiosity and sound decision-making.
That is a much more interesting future than simply producing board papers faster.
