APRA and ASIC's coordinated approach to frontier AI: implications for financial services entities

In the space of four months, APRA and ASIC have together issued public letters to industry, convened nine ‘Frontier AI Roundtables’ and published an information paper setting out their shared expectations about frontier AI. 

The message is clear: frontier AI is compounding existing cyber vulnerabilities at a pace that requires boards to engage directly with the risk, rather than treating it as a matter for operational or technology teams alone.

The regulators have also been explicit about where the burden now sits. As the joint paper puts it, for boards and executives “the question is no longer whether entities understand frontier AI risk. It is whether their entities can make decisions, maintain critical operations and recover when incidents unfold faster, affect more parties and involve more complex dependencies.” The task has moved from awareness to evidence of implementation.

A coordinated regulatory posture

The sequencing of regulatory action is instructive. APRA’s letter on 30 April 2026 (APRA Letter) drew on a targeted engagement with a sample of the largest banks, insurers and superannuation trustees, concluding that governance, risk management, assurance and operational practices are “failing to keep pace with the scale, speed, and complexity of AI adoption.” ASIC followed on 8 May 2026 with an open letter issued by Commissioner Simone Constant (ASIC Letter). Its tone is more urgent. Commissioner Constant said that “the clock is at a minute to midnight”, framing cyber resilience as “a core licensing obligation, not simply an IT issue.” For further detail on the APRA Letter and ASIC Letter, please see our previous Insight article: AI governance: ASIC and APRA letters to industry on emerging AI risks.

In June and July 2026, the two regulators convened nine roundtables across their financial services cohorts, comprising more than 380 entities and 600 attendees, including non-regulated material service providers and industry associations representing over 70,000 members. The roundtables had the support of the Australian Signals Directorate (ASD), with the RBA, Treasury and the ACCC also participating. The resulting joint information paper, Insights from the APRA-ASIC Industry Roundtables (Frontier AI Paper), was published on 27 August 2026 and consolidates the regulators’ positions into a single statement of expectations.

The combined effect of these regulatory actions represents a meaningful step-up in supervisory intensity.

Board-level expectations

At a minimum, APRA expects boards to maintain sufficient AI literacy to set strategic direction and challenge management, and to oversee an AI strategy aligned to risk appetite, with monitoring and reporting across third-party dependencies and clear triggers tied to resilience objectives.

APRA outlined its concerns about boards relying too heavily on vendor-supplied materials without independently interrogating the risks specific to AI. The practical implication is that boards will need to move beyond periodic management updates towards a substantive understanding of how AI systems operate, where they can degrade and what risks they introduce to critical operations.

ASIC has reinforced this in a more directive way, requiring entities to table the ASIC Letter at their ultimate board and risk governance committees. That converts a general warning into a documented governance obligation with an audit trail.

The general law overlay: directors’ duties and AI reliance

The expectations of regulators sit alongside (and reinforce) directors’ general law obligations. Under section 180(1) of the Corporations Act 2001 (Cth), directors must exercise the degree of care and diligence that a reasonable person would exercise in the circumstances. As the Federal Court observed in ASIC v Bekier [2026] FCA 196 (Star), that standard applies to non-financial risk oversight, including technology and operational risk governance. The decision confirms that the duty of care operates in two settings: where directors rely on AI in making decisions, and where AI itself must be governed as a source of material risk. In both, the question is whether the organisation’s information architecture gives the board adequate visibility of AI within the organisation.

Those two settings produce specific implications for boards engaging with AI:

1. AI outputs cannot constitute the decision. Directors are entitled to rely on information and advice from others, but they cannot be passive recipients. They must independently assess the information and the competence and reliability of its source. AI outputs present a particular challenge: they often cannot be fully explained, verified or challenged in the way that human advice can. AI may inform a decision, but it should not be treated as the decision itself.

2. The business judgment rule is unlikely to protect blind reliance on AI. Under section 180(2), a director’s judgment is taken to satisfy the duty of care if the director acts in good faith, for a proper purpose, without a material personal interest, is properly informed and rationally believes the judgment is in the company’s best interests. The ‘properly informed’ limb requires an inquiring mind, proportionate reliance and appropriate documentation (ASIC v Rich). A director who relies on an AI-generated recommendation without understanding its basis, limitations or assumptions is unlikely to be able to demonstrate that a ‘business judgment’ was exercised at all.

For superannuation trustees, the position is more onerous. The director covenants in the Superannuation Industry (Supervision) Act 1993 (Cth) require the degree of care, skill and diligence that a prudent superannuation entity director would exercise. This is a higher standard than the Corporations Act’s reasonable person test.

These principles underscore the regulatory message: AI literacy is not optional for boards. It is a precondition to the lawful discharge of directors’ duties where AI increasingly informs or automates decisions affecting a company’s stakeholders and critical operations.

Existing prudential standards already apply

 The regulators' output goes beyond soft guidance awaiting a future policy response. APRA has been explicit that its prudential framework is “technology and vendor agnostic”, and the Frontier AI Paper directs entities to two standards that already bind APRA-regulated entities:

1. CPS 230 Operational Risk Management, whose requirements concerning critical operations, tolerance levels, material service providers and tested exit arrangements map directly onto the third-party and concentration concerns discussed below; and

2. CPS 234 Information Security, which requires information security capability commensurate with the size and extent of threats, a threshold that moves as the threat environment does.

For APRA-regulated entities, the practical question is whether existing CPS 230 and CPS 234 compliance positions still hold once AI is added to the dependency map and the threat environment accelerates, with board obligations under CPS 510 Governance and individual accountability under the Financial Accountability Regime continuing to apply.

How frontier AI changes the risk profile

Both regulators and roundtable participants identified a consistent theme: frontier AI largely compresses timelines and amplifies existing attack surfaces rather than creating wholly new categories of cyber risk, in ways that may outpace conventional response and remediation cycles.

The Frontier AI Paper also identifies the “potential for unintended actions and outcomes from increasingly autonomous agentic systems”, a failure mode that does not require a malicious actor at all, and the one boards are least likely to have considered. 

The roundtables also highlighted a systemic dimension. Common dependence on a narrow set of third-party providers (i.e. cloud, software as a service, managed service providers, AI model providers, open-source components, payments infrastructure and telecommunications) means that an incident affecting one provider could propagate across multiple entities. Entities are directed to focus first on “shared dependencies and disruption scenarios where one weak link could affect customers, markets or confidence across the system.”

Key governance priorities

The combined regulatory output suggests five areas that boards and general counsel should prioritise:

1. Board AI literacy is now an expectation, not an aspiration. Directors who cannot challenge or justify AI-specific risks and decision-making face accountability exposure, particularly given obligations under CPS 510 Governance and the Financial Accountability Regime.
    
2. Governance must be operationalised. The APRA Letter observed a tendency to treat AI risk as “just another technology”- missing the distinct characteristics of adaptive and probabilistic systems. At a minimum, boards should expect to see: a register of AI tools and use cases, clear ownership across the AI lifecycle, defined points of human intervention for higher-risk decisions and targeted staff education on responsible use.
    
3. Cyber fundamentals are non-negotiable. The Frontier AI Paper is clear that defensive AI “is not a substitute for strong cyber fundamentals,” and that poorly governed defensive AI introduces new risks of its own. Entities must demonstrate that patching, privileged access management, monitoring, backup integrity and recovery arrangements work under compressed timeframes, especially given frontier AI enables a faster and more volatile threat environment. 
    
4. Third-party concentration is a board-level risk. The APRA Letter found entities relying on a single AI provider across multiple use cases with limited evidence that exit or substitution arrangements had been tested. Contracts often lacked protections around auditability, model changes, performance deviations, incident reporting and data handling. For APRA-regulated entities, this is CPS 230 territory. Boards should ask whether the entity could recover if a critical AI provider failed, and whether that has been tested rather than assumed.
    
5. Incident response must be pre-decided. Board decisions and guardrails (risk appetite, escalation authority, shutdown rights, supplier reliance, recovery priorities and communications) must be settled ahead of a crisis. Under compressed AI-enabled attack timelines there may be no opportunity to escalate decisions or resolve competing priorities. The Frontier AI Paper puts it directly: “if key decisions would be debated for the first time during an incident, this should be treated as a priority gap.”

Collaboration carries competition law risk that requires proactive management

The Frontier AI Paper also treats collaboration as part of resilience, urging entities to contribute to sector-wide threat intelligence sharing, dependency mapping and incident coordination. However, regulatory encouragement to collaborate does not displace market participants’ obligations to comply with the Competition and Consumer Act 2010 (Cth). The regulators themselves note that entities wishing to pursue arrangements involving “sharing of commercially sensitive information, or other forms of coordination between them that may raise competition law risks, should seek an exemption from competition laws by lodging an authorisation application with the ACCC.” However, there are other options to manage competition law risk (provided that detailed legal advice is sought and obtained), including using information sharing protocols or carefully structured joint venturing arrangements to limit problematic conduct or information transfers. 

Enforcement context

This guidance is accompanied by a clear enforcement signal. ASIC points to its court outcome against FIIG Securities Limited (26-021MR) as reinforcing “the legal case for cyber risk management controls to be demonstrably effective and proportionate to the size, nature and complexity of a business,” (see our previous Insight article for further detail). APRA has stated it will pursue enforcement where entities fail to manage AI risks proportionate to their size, scale and complexity, and is finalising a forward supervisory plan covering entity prudential reviews, thematic activities and AI supplier engagement. Both regulators have committed to continued cross-agency coordination to promote consistent expectations across the financial system.

For boards, the key question is whether the entity can demonstrate to its supervisor, and if necessary in enforcement proceedings, that it took proportionate action to address frontier AI risk in advance of an incident.

What entities should do now

1. Table the APRA Letter and ASIC Letter at the ultimate board and risk governance  committee, and record the discussion and any resulting actions.
    
2. For APRA-regulated entities, test existing CPS 230 and CPS 234 compliance positions against the entity’s current AI dependencies and use cases, including material service providers with embedded AI capability.
    
3. Run a pre-decision exercise covering escalation authority, shutdown rights, recovery priorities and communications, and document the outcomes as board-approved guardrails.
    
4. Obtain competition law advice before joining or establishing an industry information-sharing arrangement. 
    
5. Review the ASD’s Frontier AI Cyber Threat Considerations for Boards of Directors, the related Australian Cyber Security Centre’s guidance and other accompanying guidance listed in the Frontier AI Paper, which both regulators direct entities to use.