Why Australian Boards Are Starting to Ask Who Governs Their AI Systems

The short answer

Australian boards that have spent the past two years asking about Essential Eight maturity are now asking an adjacent, harder question: who is accountable for the AI tools, agents and vendor platforms entering the business? There is no single Australian AI law to point to for an answer, so directors are turning to existing frameworks — the NIST AI Risk Management Framework, APRA's prudential standards, and OAIC privacy guidance — and asking management to show evidence against them. This is becoming a governance discipline in its own right, separate from general cybersecurity assurance, because AI introduces questions of accountability, data provenance and automated decision-making that a patching schedule or a firewall review was never designed to answer.

Why the question has changed

Cyber assurance conversations in Australian boardrooms have matured quickly. Directors learned to ask about patching cadence, multi-factor authentication and backup testing because the Australian Cyber Security Centre gave them a common vocabulary for it. The Essential Eight is a set of eight fundamental cybersecurity mitigation strategies recommended by the Australian Cyber Security Centre to help organisations protect against common cyber threats, and for most private-sector boards it remains strongly recommended but not legally compulsory.

AI has arrived faster than that vocabulary could extend to cover it. Generative AI tools, copilots and now agentic systems are being procured as software, embedded in vendor platforms, and used informally by staff and even by directors themselves — often without a register, a named owner or a board-approved policy. Regulators have noticed. Both APRA and ASIC wrote to the entities they supervise this year on exactly this gap, and the first Australian court judgment to address AI use by directors has already been handed down.

What regulators and frameworks expect

NIST AI Risk Management Framework

The NIST AI Risk Management Framework has become the most widely referenced starting point for Australian organisations building an AI governance structure, even though it is a United States document with no local legal force. The NIST AI Risk Management Framework, released 26 January 2023, is a voluntary framework helping organisations identify, assess, and manage AI-related risks across four core functions: Govern, Map, Measure, and Manage. Importantly, it supports organisations that design, develop, deploy, or use AI, regardless of size or sector, which is why it translates well to an Australian mid-market business using third-party AI tools rather than building its own models.

The Govern function is the one boards should care about most directly. Govern establishes policies and accountability for AI risk management, and the Govern function emphasises the cultivation of a risk-aware organisational culture, recognising that effective AI risk management begins with leadership commitment and clear governance structures. A board that cannot name who owns that function inside the organisation has, by the framework's own definition, not yet started.

APRA CPS 234 and the 2026 AI letter

For APRA-regulated banks, insurers and superannuation trustees, CPS 234 already sets the baseline, and its logic extends naturally to AI. The Board of an APRA-regulated entity is ultimately responsible for ensuring that the entity maintains its information security, and that responsibility is not limited to internally built systems. CPS 234 requires that an APRA-regulated entity take necessary measures to defend itself from information security incidents concerning the confidentiality, integrity and availability of information assets and data — this includes information managed by third-party service providers, reflecting an increased regulatory focus on third-party risk. Every AI vendor platform handling customer data falls squarely inside that obligation.

APRA has since made the AI connection explicit. In April 2026, APRA wrote to regulated entities with findings from a targeted engagement with large banks, insurers and superannuation trustees, and named weaknesses across cyber and information security, governance, supplier risk, and change management and assurance. The letter also set an expectation directed squarely at the board table: directors must hold sufficient understanding and literacy to provide effective challenge on AI risk and strategic direction. Shortly after, ASIC issued an open letter on 8 May 2026 on cyber resilience in the face of frontier AI, and asked that it be tabled and discussed at board and risk committee level.

Directors' duties and the courts

Existing Australian directors' duties have not changed because AI arrived — they simply now apply to a new category of risk. Legal commentary on recent case law suggests directors must consider and approve an AI governance framework that covers approved tools, uses, accountability and distinctions between low- and high-risk applications, develop enough AI literacy to challenge management and oversee risk effectively, and maintain independent judgement when relying on AI-assisted information. The first Australian judgment to consider AI use by directors reinforced that boards should treat AI use as something to be deliberately governed through the formal adoption of policies, rather than permitting informal "shadow" use to go unchecked.

OAIC and the Privacy Act

Where AI systems touch personal information, the OAIC's existing guidance already sets clear expectations. As a matter of best practice, the OAIC recommends that organisations do not enter personal information, and particularly sensitive information, into publicly available generative AI tools, due to the significant and complex privacy risks involved. The regulator has also confirmed Privacy Impact Assessments should be performed before a new AI system is introduced, and expects that an APP entity's practices, procedures, systems and privacy policy account for its handling of personal information — including through AI tools.

Comparing what each framework actually asks a board to show

← Swipe to compare →

The questions boards are starting to ask

The questions appearing in Australian board papers are converging on a consistent set. Directors should ask which agentic systems are live, what data they can access, what actions they can take, who approved them and what controls apply.

• Who is the single named owner of AI governance, equivalent to how we name an owner for finance or safety risk?
• Do we have a register of AI tools and automated-decision systems in use, including tools staff have adopted informally?
• Have our AI vendor contracts been reviewed for data use, transparency and audit rights, the way other third-party contracts are?
• Is AI risk reported to the board on a defined cadence, with clear triggers for escalation?
• Has a Privacy Impact Assessment been completed before each new AI system touching customer or staff data went live?

How to prepare — practical steps

Boards and security leads do not need to wait for new legislation to close this gap. A defensible starting position can be built in a single quarter by acting on a short list of fundamentals: name a single accountable executive for AI, the way you would for finance or safety; build and maintain an AI and automated-decision use-case register, including shadow AI; put AI and data clauses into vendor and managed service provider contracts, covering transparency, data use, and your right to audit; and report AI to the board on a regular cadence, with defined triggers for when management must escalate.

Alongside that register, it is worth running a technical baseline check against the Essential Eight for any infrastructure hosting or connecting to AI tools, and a privacy review against OAIC guidance for any tool that touches personal information. Where the gap is less about AI strategy and more about there being no one in a security leadership seat at all, a fractional CISO (/grc/vciso/) is often the faster fix than recruiting a full-time hire.

Why this is becoming a distinct discipline

General cybersecurity assurance answers questions about confidentiality, integrity and availability of systems. AI governance adds questions those controls were never built to answer: whether an output can be explained, whether a vendor's model was trained on data it should not have had, whether an agentic tool took an action nobody approved, and whether the organisation can show a regulator it assessed these risks before deployment rather than after an incident. That is why AI oversight increasingly sits as its own line on the board agenda, with its own owner, rather than being folded into the existing cyber report.

How Infracom helps

Infracom governs, assesses and secures AI systems for Australian and Singapore organisations — we do not build, develop or train them. Our consultants hold AAISM credentials and work within an ISO 27001-certified management system, and our engagements include AI governance gap assessments against the NIST AI RMF, vendor AI risk reviews aligned to APRA CPS 234 and OAIC guidance, and structured Essential Eight advisory for the infrastructure AI tools sit on. For boards that need a named, accountable owner of AI governance without a full-time executive hire, Infracom's fractional Chief AI Officer service (/grc/vcaio/) provides that oversight on a scale matched to mid-market and regulated organisations.