This article was co-authored with Amy Moore.
Artificial intelligence (AI) is reshaping how Australian financial institutions detect fraud, manage risk, and interact with customers. Yet the promise of faster, data-driven decision-making comes with stringent regulatory expectations. Four key regulators – the Australian Securities and Investments Commission (ASIC), the Australian Prudential Regulation Authority (APRA), the Office of the Information Commissioner (OAIC), and the Australian Transaction Reports and Analysis Centre (AUSTRAC) – already oversee many facets of AI in the financial services sector. The majority of these regulators have made clear that existing obligations on financial services providers apply with full force regardless of whether AI tools are deployed.
To ensure compliance in practice, consideration should be given to AI-specific concerns such as opacity, complexity and the ability to generate personal information, among other things. Boards, executives and compliance teams need to be considering how they will evidence compliance when algorithms sit at the heart of core processes. This article explains what Australian financial services providers need to know, offering practical compliance tips and outlining key questions directors should ask about AI deployment.
Why AI matters to financial regulators
AI is already deeply embedded in the Australian financial services sector. Common use-cases include transaction monitoring for suspicious activities; fraud, and abuse; credit scoring and lending decisions; customer onboarding and identity verification; robo-advice and personalised product recommendations; automated customer service via chatbots and voicebots; and stress testing or portfolio risk management. Machine learning models analyse transaction patterns to flag potential money laundering, automate creditworthiness assessments for loan applicants, streamline customer identity verification during onboarding, and provide tailored investment advice or support via digital assistants.
These and other AI-driven functions intersect squarely with regulators’ mandates of protecting market integrity, ensuring fair customer outcomes, detecting financial crime and safeguarding personal information. Yet while machine learning models can enhance these objectives, they may also introduce opacity, bias and systemic risk. Opacity arises when complex algorithms' decision-making processes are difficult for firms and regulators to interpret or audit. Bias can be embedded in training data or model design, potentially leading to unfair or discriminatory outcomes for certain customer groups. Systemic risk may increase if multiple institutions rely on similar AI models, amplifying the impact of errors or vulnerabilities across the sector. As AI systems become more deeply embedded in critical financial processes, both regulators and financial services providers face the challenge of interpreting and adapting compliance programs and oversight frameworks to address new forms of risk and complexity.
Regulator guidance on AI and financial services obligations
Australian financial services regulators favour a 'technology-neutral' stance on AI: they neither exempt AI from existing rules, nor see an immediate need for bespoke legislation. Instead, the expectation is that relevant organisations will adapt existing compliance frameworks to AI tools' unique characteristics. The implications of this approach are significant but navigable, as outlined in this article.
ASIC
ASIC focuses on ensuring compliance with the Corporations Act 2001 (Cth) and ASIC Act 2001 (Cth), irrespective of whether decisions are human or algorithmic. In guidelines released in 2024, ASIC expressed concerns about a governance gap between AI adoption and risk management. It encouraged financial services providers to proactively comply with existing (technology-neutral) obligations when adopting AI. For example:
- AI bias and opacity, and representations about a system's error rates, require careful consideration against the obligation to provide financial services efficiently, honestly, and fairly under section 912A of the Corporations Act, and the prohibition on misleading or deceptive conduct under section 12DA of the ASIC Act and section 1041H of the Corporations Act.
- The inferential power of AI must not exploit vulnerable consumers (e.g., by predicting vulnerability to high-interest loans), as this would violate the prohibition on unconscionable conduct under sections 12CB-CC of the ASIC Act.
- Directors’ duties of care and diligence are applicable both if deploying AI in the business, and in using AI in one’s own role as a director.
ASIC further encourages providers to establish an AI strategy aligned with business objectives and ethical principles; create internal committees or roles for AI oversight; ensure sufficient human supervision (including compliance, risk, and business expertise); effectively manage third-party provider risks; and document and review AI systems and updates. ASIC’s 2025–26 Corporate Plan places a strong focus on enhancing AI oversight and strengthening cyber security within regulated organisations, underlining the need to ensure that technological advancements are implemented in a manner that is safe, ethical and responsible.
ASIC has also stated that it expects financial services providers to prioritise cyber resilience and invest in systems and governance that are fit for purpose and proportionate to their size and sensitivity of the information they hold. This necessarily extends to AI governance and the data it accesses, trains on and generates.
The need for robust governance of systems was reinforced by the Federal Court’s decision to impose $2.5 million in pecuniary penalties against FIIG Securities Limited (FIIG),1 after ASIC brought regulatory action against FIIG for cybersecurity deficiencies that enabled the theft of approximately 385GB of sensitive client data, affecting around 18,000 clients. This was the first time penalties have been ordered for cyber security failures under the general Australian Financial Services licensee obligations, underscoring the importance of implementing and maintaining well-governed systems - an expectation that is increasingly significant with the growing adoption and use of AI as well as the potential for AI to increase cybersecurity risk.
APRA
Whilst APRA, the Australian prudential regulator for banks, insurers and superannuation funds, has not issued AI-specific rules, its existing prudential standards already address key risks posed by AI. These standards require firms to integrate AI systems into their established risk management, information security and operational resilience frameworks. Two standards are particularly central:
- Prudential Standard CPS 234 (Information Security) demands controls that protect the confidentiality, integrity and availability of information assets. AI models, training data and inferences all fall within scope; firms must secure them against unauthorised access or manipulation that could corrupt outputs or place systems at risk.
- Prudential Standard CPS 230 (Operational Risk Management) hard-wires expectations around resilience, noting that complex AI supply chains – cloud providers, data vendors, etcetera – can be points of failure. Regulated businesses need to ensure their contracts with AI vendors are updated to be CPS 230-compliant by the next renewal and at the latest by 1 July 2026.
For 2026, APRA is placing a strong emphasis on enhancing cyber resilience, particularly by keeping a close watch on new risks that may arise as artificial intelligence is adopted. The regulator will work with major regulated organisations to evaluate if their approaches to risk management and governance are sufficient to ensure AI is integrated responsibly throughout the financial sector.2
OAIC
While the OAIC is not a financial services regulator as such, its administration of the Privacy Act 1988 (Cth) and the Australian Privacy Principles (APPs) is central to the governance of AI systems used by financial services to handle personal information. Most financial services providers fall within the scope of the Privacy Act, typically having an annual turnover exceeding $3 million, and must therefore ensure that AI adoption complies with APP obligations, including in relation to transparency, necessity, notification, limited use and disclosure, security, quality and accuracy.
The OAIC has issued guidance to assist entities in understanding and meeting their privacy obligations when adopting and deploying AI. The guidance is divided into two parts: the first guide is for businesses using commercially available AI products and the second guide relates to use of personal information to develop and train generative AI. Both guides emphasise the need for caution when adopting AI, due to the significant privacy risks and unique compliance challenges that come with it. In particular:
- AI can use, infer and generate personal information that users have not knowingly provided.
- The opacity of complex algorithms can make it difficult for entities to meet APP requirements relating to transparency, particularly when describing how personal information is used but will require transparency in respect of the personal information used in automated decision-making by December 2026.
The OAIC encourages regulated entities to implement measures to mitigate risks, including the following:
- Maintain strong internal policies for both the procurement and use of AI. This includes conducting due diligence to ensure AI tools are fit for purpose and do not pose security risks. Once implemented, users should receive appropriate training on how to safely use the product, including guidance on the types of information that should not be input into AI tools.
- Conduct a privacy impact assessment before implementing new AI tools.
- Ensure ongoing human oversight and accountability supported by governance processes that continuously monitor AI use and outputs, and ongoing staff training in relation to AI risks and limitations.
AUSTRAC
Under the Anti-Money Laundering and Counter-Terrorism Financing Act 2006 (Cth), reporting entities (which include certain businesses in the financial sector) must maintain a compliant anti-money laundering and counter-terrorism financing (AML/CTF) program, run transaction-monitoring systems, file suspicious matter reports, and conduct ongoing customer due diligence. AUSTRAC, the AML/CTF regulator, has not issued specific guidance on AI. However, if AUSTRAC follows the other financial services regulators in taking a technology-neutral stance, then financial services providers considering the adoption of AI to assist with their AML/CTF programs would be prudent to consider taking at least the following steps:
- Updating their written AML/CTF programs to describe any AI-driven monitoring or screening tools, including legacy systems.
- Ensuring that algorithms genuinely enhance detection capabilities rather than create “blind spots” – for example, by over-fitting to historical patterns that exclude new money laundering patterns.
- Ensuring they are able to explain the logic of an AI system, for example, why the system flagged certain instances as fraudulent and not others.
- Retaining qualified human oversight, ensuring AI is a tool, not a substitute for risk-based judgment or mandatory reporting. Safeguards should always be proportionate to risk.
Providers should also be aware of changes to the AML/CTF regime (including an expansion of the sectors covered by the regime), which will come into effect in stages throughout the first half of 2026.
Practical steps to close the AI compliance gap
Drawing together existing regulator messaging on AI, five concrete actions are recommended for financial services providers.
- Integrate AI into existing compliance programs: For AUSTRAC obligations, this means rewriting AML/CTF programs to describe how AI models are trained, validated and calibrated.
- Treat AI as a tool, not a replacement: Human oversight remains mandatory. Establish escalation thresholds so anomalous AI outputs, such as unusually low fraud scores, trigger manual review. Keep a documented rationale for human overrides, providing evidence of reasonableness.
- Update vendor contracts: Most AI solutions involve third-party software or data. Contracts must now cover algorithmic transparency, audit rights, incident notification and model-change governance. Allocate liability for privacy breaches or erroneous outputs and require vendors to maintain minimum APRA-style security controls.
- Develop AI-specific incident response plans: Traditional cyber playbooks assume a data breach or service outage. AI failures can also include model corruption, adversarial attacks or discriminatory outcomes. Plans should include immediate containment of erroneous models, communication protocols to customers and regulators, and rapid retraining or rollback procedures.
- Refresh privacy notices and consents: From December 2026, privacy legislation will mandate explicit statements about automated decision-making. Financial institutions should act early: explain in customer-facing language what data is used, the logic of decisions (at a high level) and the potential impacts. Where possible, build opt-out or review pathways for significant automated decisions. Regular reviews of the currency of those statements will be an on-going task for organisations as their use of AI changes and adapts.
Governance considerations for boards and executives
Boards ultimately sign-off on risk-management frameworks under CPS 220, bear accountability for AML/CTF programs and attract liability under the Corporations Act. To assist with AI compliance, boards should be asking management questions such as:
- Where in our value chain do AI models operate, and who owns them?
- How are model risks such as bias, data drift, and false positives and negatives monitored and reported?
- Do we have documented explanations sufficient to satisfy regulators if challenged?
- What evidence shows our AML/CTF models outperform or equal traditional rules-based systems?
- Which third parties touch our data or model pipelines, and do contracts embed APRA-grade controls?
Looking ahead: Regulatory trajectory
While regulators currently apply existing laws in a technology-neutral manner, more tailored rules for AI may be on the way. The Australian Government has already introduced a Voluntary AI Safety Standard comprised of ten guardrails. International developments such as the EU AI Act influence Australian approaches and may prompt sectoral guidance. Nonetheless, the core message is unlikely to change – obligations of privacy, fairness, security and risk management are technology-neutral, and non-compliance with them, even when unintended, attracts the same enforcement consequences.
Conclusion
Australian financial services firms are well-placed to harness AI’s benefits, from sharper fraud detection to enriched customer experiences. But success depends on embedding AI within established compliance disciplines, not bolting it on afterwards. By updating programs, notices, contracts and incident playbooks now, financial institutions can demonstrate to the full range of relevant regulators and their customers that AI is deployed responsibly, thoughtfully and in alignment with Australia’s existing legal framework and its layers of regulation. In doing so, they will not only meet compliance hurdles but also build the trust essential for sustained innovation in an AI-driven financial future.
Footnotes
-
Australian Securities and Investments Commission v FIIG Securities Limited [2026] FCA 92.
-
Australian Prudential Regulation Authority, ‘APRA Corporate Plan 2025-26’ <https://www.apra.gov.au/apra-corporate-plan-2025-26>.