Why AI board oversight has become urgent
Boards have traditionally treated technology risk as something management handles and reports up. AI is breaking that model. The speed at which AI tools are being adopted, the novelty of the threats they introduce, and the direct exposure they create at the executive level all mean that waiting for a management summary is too slow.
Consider what has changed in the past two years. Generative AI tools are now available to every employee with a credit card and a browser. A finance officer can upload a supplier contract to summarise it. A sales director can paste a client proposal into a chatbot for feedback. Neither action requires IT approval, and neither leaves a trace in your security logs. The data is gone before anyone knows to ask.
At the same time, the fraud environment has shifted. Voice cloning and deepfake video are no longer expensive or technically demanding. They are accessible to anyone motivated to commit fraud, and Australian organisations have already been targeted. A convincing audio clip of a CEO authorising a payment is not a theoretical risk.
Regulatory pressure is also moving. Australian privacy law is being updated, and guidance from regulators including ASIC and APRA is signalling that boards will be expected to demonstrate active oversight of AI risk, not merely awareness of it. Ignorance of how AI is being used inside your organisation is unlikely to be a defensible position.
The four risks below are not exhaustive. They are the ones most likely to surface as a material issue before the end of your current financial year.
Risk one: fraud and impersonation at executive scale
AI has made it cheap and fast to impersonate a person convincingly. A voice clone of a CFO can be produced from a few minutes of publicly available audio. A realistic video of a CEO authorising a transaction can be generated without any footage of them ever saying it. These are not theoretical capabilities sitting in a research lab. They are being used in financial fraud right now, and Australian organisations are among the targets.
The evolved form of business email compromise (BEC) is worth naming precisely. Traditional BEC involved a criminal sending an email that appeared to come from an executive, typically requesting an urgent funds transfer. The receiving employee had one signal to evaluate: does this email look legitimate? AI-assisted fraud adds voice, video and real-time conversation to that attack surface. A finance team member receives a call that sounds exactly like the CFO. A payment is approved. The money is gone before anyone checks.
Boards should understand that the controls designed for email-based fraud do not automatically extend to voice and video channels. A callback policy that routes verification through a known number costs almost nothing to implement and stops a significant portion of these attacks. The question is whether anyone in the organisation has been trained to use it under pressure.
There are two questions a board should be asking directly. First, has the organisation updated its payment authorisation and funds transfer protocols to account for AI-generated voice and video? Second, have the staff most exposed to these requests, specifically finance, legal and executive assistants, received specific training on what to do when something feels wrong?
The CFO's guide to AI deepfake fraud goes deeper on the mechanics and the internal controls worth prioritising. The short version: the strongest defence is not technology. It is staff who know what to verify, how to verify it, and who feel safe slowing down a request even when it appears to come from a senior person.
Risk two: data leakage through everyday AI tools
Most data breaches boards think about involve an external attacker. The more common AI-era risk is quieter: a staff member pastes a client contract into ChatGPT to get a summary, or uploads a financial model to an AI assistant to clean up the formatting. No malice, no hacking, just a reasonable person trying to get their work done faster.
The problem is where that data goes. Consumer AI tools, and many enterprise ones if not configured correctly, may use submitted content to train future models, store it on servers outside Australia, or expose it to other users through system vulnerabilities. A staff member who would never email a client's personal information to a personal account will do the functional equivalent without a second thought if the interface looks like a productivity tool.
The questions worth asking at board level are straightforward. Does the organisation have a clear policy on what categories of data may be submitted to external AI tools? Has that policy been communicated, not just published? And does IT have visibility into which AI tools are being used across the business?
That last question is harder to answer than it sounds. Research into enterprise AI use consistently finds that the tools employees use day-to-day extend well beyond what IT has approved or is even aware of, a pattern explored in depth in the companion piece on shadow AI and unmanaged tool proliferation. The practical consequence is that your data classification policy may be sound on paper while being largely invisible to the people whose behaviour it is meant to govern.
For regulated industries, the stakes are higher still. Uploading patient records, personal financial information, or legally privileged documents to an unsanctioned AI tool can trigger obligations under the Privacy Act, the Notifiable Data Breaches scheme, and sector-specific regulation. The Privacy Act and AI article covers the updated requirements in detail. Boards in financial services, healthcare, and legal services should be asking whether their AI tool policies specifically address these categories, and whether training reinforces the boundaries in practice rather than just at induction.
The remediation here is not a blanket ban on AI tools. That tends to push usage further underground. A more effective approach combines a clear acceptable use policy, tool-level configuration (enterprise licences for approved tools typically include data residency and training-opt-out controls), and regular awareness training that gives staff a concrete mental model of what the risk actually looks like.
Risk three: regulatory and compliance exposure
Australian AI regulation is moving faster than most boards realise. The Privacy Act amendments introduced stricter obligations around automated decision-making and the handling of personal information, and the Australian Government's voluntary AI Safety Standard signals where mandatory requirements are likely to head. Boards that are waiting for a single definitive AI law before acting are misreading the landscape. Obligations already exist, scattered across privacy law, anti-discrimination legislation, sector-specific regulation, and consumer protection rules.
The compliance risk is not abstract. If an AI tool used in recruitment screens out candidates in ways that correlate with a protected attribute, the organisation can face action under the Sex Discrimination Act or the Disability Discrimination Act regardless of whether a human reviewed the outcome. If a customer-facing AI tool retains or processes personal information in ways the privacy policy does not disclose, that is a potential notifiable data breach or an interference with privacy under the Privacy Act 1988. The board does not need to understand the technical architecture to be accountable for either outcome.
Three specific areas deserve a standing question on every board agenda.
Automated decisions affecting individuals. Any AI system that influences hiring, credit, insurance, or access to services sits in a high-risk category. The AI in HR and recruitment bias and compliance traps article covers this in detail, but the board-level question is simple: do we know which decisions in this organisation are being influenced or made by AI, and have we assessed whether that creates discrimination or privacy exposure?
Data residency and third-party AI tools. When staff use a cloud-based AI tool, personal information often flows to servers outside Australia. The Privacy Act's Australian Privacy Principle 8 governs cross-border disclosure. Many commercial AI tools in common use were not procured with that obligation in mind. The board should be asking whether the organisation's AI tool procurement process includes a privacy impact assessment, not just a security review.
Disclosure and transparency obligations. Several regulators, including ASIC and the ACCC, have signalled expectations around transparency when AI influences outcomes that affect consumers or investors. If your organisation uses AI in any customer-facing or market-sensitive process and has not reviewed its disclosure obligations, that gap is a board-level risk.
The honest challenge for most boards is that these questions require someone in the organisation who understands both the regulatory environment and the actual AI tools in use. That combination is rare. The Australian AI regulation in 2026 article in this series covers the specific legislative developments in more detail. The governance implication for boards is that waiting for legal counsel alone is not sufficient: counsel can interpret the law, but they need accurate information about what the AI systems are actually doing to advise on exposure.
Risk four: shadow AI and unmanaged tool proliferation
Shadow AI is what happens when employees adopt AI tools without IT, legal, or management knowing about it. It is not a theoretical concern. When a productivity tool is free, instant, and genuinely useful, people use it. A marketing manager pastes a draft strategy into ChatGPT. A finance analyst uploads a spreadsheet to an AI summarisation tool. A lawyer runs a contract through a browser-based AI reviewer. None of these people are trying to create a problem. They are trying to get their work done faster.
The governance blind spot this creates for boards is significant. If a tool was never reviewed, it was never assessed for data handling practices, jurisdiction of data storage, or terms of service. The organisation may have no idea what data has left its systems, which vendors hold it, or how those vendors use it to train future models.
Shadow AI tends to proliferate fastest in organisations that have been slow to provide approved alternatives. When there is no clear answer to "which AI tool should I use for this?", people find their own. The gap between what IT sanctions and what employees actually need is where shadow AI lives.
For boards, the questions worth asking are direct. Does the organisation have an acceptable use policy that names specific approved tools? Is there a lightweight process for employees to request a new tool rather than just use it? And does the CISO or CIO have any visibility into which AI services are being accessed on corporate networks?
These are not questions about slowing down AI adoption. Organisations that establish clear guardrails often see faster, more confident adoption than those where employees are uncertain about what is and is not acceptable. Clarity accelerates use; ambiguity creates both hesitation and risk.
The shadow AI problem is closely connected to the data leakage scenarios described in risk two. Unmanaged tools and uncontrolled data flows compound each other. Addressing them requires visibility first, policy second, and training to make the policy stick.
What good AI board oversight actually looks like
Oversight does not mean directors become AI experts. It means boards ask the right questions and expect credible, specific answers, not reassuring generalities.
Here is what that looks like in practice.
Ask for an AI inventory, not a strategy summary. Management should be able to tell the board which AI tools are in use, who approved them, and what data each one can access. A strategy document is useful; an actual list is evidence. If the list does not exist, that is the first finding.
Require a named owner for AI risk. In many organisations, AI risk sits in a gap between IT, legal, and the executive team, with nobody formally responsible for the whole picture. The board should ask: who is accountable when something goes wrong? That person's name should be on a slide, not implied by an org chart.
Test the incident response, not just the policy. Most organisations now have an AI policy. Far fewer have tested what happens when a deepfake impersonating the CFO is used to authorise a payment, or when a staff member pastes a client contract into a public AI tool. Ask management to walk you through the last time a scenario like this was rehearsed. The quality of that answer tells you a great deal.
Request a training baseline, not a training plan. There is a meaningful difference between "we are planning to roll out AI training" and "here is the percentage of staff who have completed it, by role." Boards should be seeing the latter. Awareness training for high-risk roles like finance, legal, and executive assistants should be tracked separately, given those are the staff most likely to be targeted by AI-enabled fraud.
Look at vendor contracts, not just vendor names. When management lists the AI tools the organisation uses, the follow-up question is: what do those contracts say about data retention and training? Consumer-grade AI tools often include clauses permitting the provider to use inputs to improve their models. Enterprise agreements typically do not, but only if the right tier has been procured and configured correctly. This is a governance detail that lives in contracts, not in marketing collateral.
Set a reporting rhythm. AI risk should appear on the board agenda at a defined cadence, at least twice a year, with a short standing update at every meeting while the organisation is actively deploying new tools. Ad hoc reporting, triggered only when something goes wrong, is a control gap. The board should be informed before incidents, not only after them.
None of these questions require technical knowledge to ask. They require the same discipline boards already apply to financial risk, WHS, and privacy: clear ownership, evidence over assurance, and a reporting structure that surfaces problems early.
Frequently asked questions
What is AI board oversight and why does it matter now?
AI board oversight means the board actively monitoring, questioning and setting accountability for how AI is used across the organisation, rather than treating it as a technical matter left entirely to management. It matters now because AI tools have moved out of IT and into everyday workflows across finance, HR, legal and operations. Boards that are not asking questions about AI risk are effectively signing off on exposures they have not seen.
How is AI risk different from the cybersecurity risk boards already govern?
AI risk overlaps with cybersecurity but is not the same thing. Cyber governance typically focuses on external attackers and system vulnerabilities; AI risk also includes decisions made by staff, the tools they choose without IT approval, and the outputs those tools generate. A deepfake voice call that tricks a finance officer into transferring funds is both a cyber event and an AI event, and the controls that would prevent it sit closer to policy and training than to firewalls.
What questions should a board actually be asking in a board meeting?
Four questions worth putting on the agenda: Which AI tools are staff currently using, including tools that were not formally approved? What customer or commercial data has been entered into those tools? Has the organisation reviewed its AI exposure under the Privacy Act and any applicable sector regulation? And who is accountable when an AI-related incident occurs? If management cannot answer these clearly, that is itself useful information.
Does the board need to understand AI technically to provide meaningful oversight?
No. Boards govern financial risk without every director being an accountant, and the same logic applies here. What directors need is a clear picture of where AI creates material risk, a management team that can articulate that picture, and enough literacy to ask the right follow-up questions. AI scam awareness training and short executive briefings are often enough to close that gap without requiring deep technical knowledge.
How often should AI risk appear on the board agenda?
At least quarterly, and more frequently during periods of rapid AI adoption or regulatory change. AI risk is not a set-and-forget item. New tools appear constantly, staff behaviour changes, and the regulatory environment in Australia is still developing. Boards that review AI governance annually are almost certainly working with outdated information.
Is your organisation's AI risk visible to the board?
Most boards are not short on intent when it comes to AI governance. What they are often short on is a clear picture of where the risks actually sit right now, before an incident forces the question.
The four risks covered here, fraud and impersonation, data leakage, regulatory exposure, and shadow AI, are not hypothetical. They are present in most Australian organisations today, often without any board-level visibility. Bringing them into the governance conversation does not require a technology overhaul. It requires the right questions, asked regularly, by people with enough context to push back on reassuring non-answers.
That is exactly the conversation a discovery call with Better People is designed to start. We work with boards, executive teams, and L&D leads to identify where AI risk is genuinely unmanaged, and what a practical response looks like for your organisation's size, sector, and current maturity.
If you want to go deeper before that conversation, the AI risk, governance and scams pillar hub brings together everything from Privacy Act obligations to verification protocols in one place.
