APRA's Expectations for AI Fraud Detection in Insurance: A Compliance Guide

The Australian Prudential Regulation Authority (APRA) has made its position clear: insurers deploying artificial intelligence must do so within their existing risk and governance frameworks. In April 2026 it wrote to every regulated entity calling for a “step-change” in AI risk management, and it named claims triage and fraud and scam disruption among the AI uses it sees in the industry. For insurers adding deepfake detection to claims, understanding APRA’s expectations is a board-level issue.

This guide maps APRA’s prudential framework to the practical questions raised by AI-powered deepfake detection in insurance claims.

The Regulatory Landscape for AI in Australian Insurance

APRA does not regulate AI through a single, standalone standard. Its expectations are spread across prudential standards and guidance that together form a framework for technology and operational risk.

The key instruments relevant to AI fraud detection include:

CPS 230 Operational Risk Management (in force since July 1, 2025)
CPS 234 Information Security
CPS 220 Risk Management
CPG 235 Managing Data Risk (guidance, not a binding standard)
CPG 230 and CPG 234 (the practice guides that accompany CPS 230 and CPS 234)
The Financial Accountability Regime (FAR), which has applied to insurers since March 15, 2025

APRA’s senior leaders have said repeatedly that the existing framework already captures AI. In a September 2025 speech, APRA Executive Director Chris Gower said the “existing regulatory framework is sufficient to capture the use of AI,” and APRA Member Suzanne Smith made the same point in October 2025, noting that the information security, operational risk, data risk and risk management standards all apply. APRA’s April 30, 2026 letter to industry on AI added no new requirements, but found that while most entities recognize existing standards apply to AI risk, few have operationalized that governance in practice.

The practical implication: insurers cannot treat AI fraud detection tools as exempt from standard governance requirements.

CPS 230: Operational Resilience and Deepfake Detection

CPS 230 replaced the earlier outsourcing and business continuity standards (CPS 231 and CPS 232, and their superannuation and health equivalents). It is the most significant change to APRA’s operational risk framework in over a decade, and it bears directly on AI-powered fraud detection.

Critical Operations

Under CPS 230, insurers must identify their critical operations — processes whose disruption would have a material adverse impact on policyholders or the entity’s role in the financial system. The standard lists claims processing as a critical operation for general, life and private health insurers unless the entity can justify otherwise. Fraud detection embedded in claims processing therefore sits inside a critical operation.

If deepfake detection is integrated into the claims workflow, the resilience requirements that attach to claims processing need to account for it. In practice this means:

Tolerance levels for claims processing — maximum disruption period, maximum data loss and minimum service levels — should consider what happens when the detection capability is unavailable
Testing of business continuity arrangements should include scenarios where AI detection tools are down
Fallback procedures (for example, manual escalation of higher-risk claims) should be defined and documented

CPS 230 also sets notification deadlines: APRA must be told within 72 hours of an operational risk incident likely to have a material financial impact or a material impact on critical operations, and within 24 hours when a disruption takes a critical operation outside its tolerance levels.

Third-Party Risk Management

CPS 230 sets requirements for material service providers — providers the entity relies on to undertake a critical operation, or that expose it to material operational risk. Insurers must treat providers of claims management services as material unless they can justify otherwise. Whether a third-party deepfake detection platform — including an API-based service like deetech™‘s — is a material service provider depends on how deeply it is embedded in claims processing.

Where it is material, CPS 230 requires, among other things:

A register of material service providers, submitted to APRA annually
A formal, legally binding agreement covering matters such as data access, APRA’s access rights and termination
Notification to APRA within 20 business days of entering into or materially changing an agreement for a critical operation
Ongoing monitoring of the provider, and the ability to exit the arrangement in an orderly way if needed

None of this discourages third-party AI tools. The requirement is that they be governed properly.

Incident Management

When a deepfake detection system produces an incorrect result — a false positive that delays a legitimate claim, or a false negative that lets a fraudulent one through — this may be an operational risk incident under CPS 230. Insurers need clear escalation protocols and root-cause analysis processes for AI system failures.

CPS 234: Information Security Implications

CPS 234 requires APRA-regulated entities to maintain information security capability commensurate with the threats they face. Deepfake fraud is itself an information security threat, and the tools used to counter it create their own security considerations.

Data Classification

Deepfake detection systems process sensitive material: images and recordings of people, identity documents and claims evidence. CPS 234 requires information assets to be classified by criticality and sensitivity. Under the Privacy Act 1988, biometric information used for automated biometric verification or identification, and biometric templates, are “sensitive information” — so insurers should also understand whether any tool in the chain performs that kind of processing.

Insurers should ensure that:

Data processed by deepfake detection tools is encrypted in transit and at rest
Access controls follow least-privilege principles
Audit trails capture access to claims media and analysis results
Data retention policies comply with both APRA requirements and the Privacy Act 1988

CPS 234 also requires material information security incidents to be notified to APRA within 72 hours, and material control weaknesses within 10 business days.

Control Testing

AI models are susceptible to adversarial attacks — deliberate attempts to evade detection. CPS 234 requires systematic testing of information security controls, and those controls extend to AI systems. Insurers should:

Regularly assess deepfake detection against emerging attack techniques
Maintain update processes for AI components
Include adversarial AI scenarios in security testing where relevant

CPG 235: Managing Data Risk in AI Systems

CPG 235 is guidance rather than a binding standard, but its data risk principles are directly applicable to AI fraud detection.

Data Quality

The effectiveness of deepfake detection depends heavily on data quality. CPG 235 describes better practice for data quality, including accuracy, completeness, timeliness and consistency. For deepfake detection, this suggests:

Model data used to build and evaluate detection models should be representative of the media the insurer actually receives
Input data (claims evidence, identity documents) should meet minimum quality thresholds before being analyzed
Output data (detection results, confidence scores) should be validated before being relied on

Data Lineage

Good data risk management includes the ability to trace data from its origin through all transformations to its final use. For deepfake detection, this means documenting:

Where claim evidence originated (e.g., uploaded by the claimant, captured in person)
What processing the evidence underwent before analysis
Which system and version produced the detection output
How that output influenced the claims decision

This lineage is critical for defending claims decisions if they are challenged by policyholders, the Australian Financial Complaints Authority or a court.

Privacy Act Changes Arriving in December 2026

The Privacy and Other Legislation Amendment Act 2024 adds automated decision-making transparency obligations that commence on December 10, 2026. From that date, entities must explain in their privacy policies when they use personal information in computer programs that make, or substantially assist in making, decisions that could significantly affect an individual’s rights or interests. Insurers using AI-assisted fraud screening in claims should assess now whether their processes fall within these provisions.

Model Risk Management

APRA has not issued a standalone model risk management standard for insurers, but its expectations run through CPS 220, CPS 230 and supervisory engagement. Good practice for deepfake detection models includes:

Model Validation

Before relying on a deepfake detection model in production, insurers should:

Evaluate performance on representative data, including the media types, devices and compression levels found in the insurer’s own claims
Stress test the model against known adversarial techniques
Benchmark on data independent of the vendor, rather than relying solely on vendor-reported metrics
Document the model’s limitations and known failure modes

Ongoing Monitoring

Deepfake technology evolves quickly. A detection model that performs well today may be less effective against next-generation synthetic media. Good practice includes:

Regular, documented performance reviews, with frequency proportionate to the model’s risk
Monitoring of flag rates, confirmed false positives and missed cases
Trigger-based reviews when the threat landscape changes significantly
Review of model performance by people independent of the team that owns the model

Model Governance

Clear accountability for AI models means:

A designated owner at senior management level (under FAR, accountability for claims and risk functions should be clearly mapped to accountable persons)
A model risk management framework covering the full lifecycle
Board reporting on material model risks, including AI fraud detection performance
Documentation sufficient for APRA to understand and assess the model during supervisory engagement

Practical Compliance Checklist

For insurers implementing or reviewing deepfake detection against APRA expectations:

Governance

Board-approved AI governance framework that covers deepfake detection
Designated senior management accountability for the AI fraud detection model
Regular board reporting on model performance and material incidents

CPS 230

Deepfake detection mapped to the claims processing critical operation
Tolerance levels and fallback procedures consider detection system disruption
Material service provider assessment completed and documented for the detection provider
Exit arrangements in place where the provider is material
Business continuity plan covers AI detection system failure scenarios

CPS 234

Information assets processed by detection systems classified by criticality and sensitivity
Encryption in transit and at rest for claims media and results
Access controls and audit trails implemented
Control testing includes adversarial AI scenarios where relevant

Model Risk

Model evaluation completed on representative data
Ongoing monitoring regime established with defined metrics and thresholds
Independent review process in place
Model documentation sufficient for supervisory review

Data and Privacy

Data quality standards defined for input and output data
Data lineage documented from evidence capture through to claims decision
Data retention and disposal policies aligned with the Privacy Act and APRA requirements
Automated decision-making disclosures reviewed ahead of December 10, 2026

APRA’s Evolving Posture on AI

APRA’s approach to AI is principles-based rather than prescriptive. It has said it has no new AI regulations planned and expects regulated entities to manage the risks of new technologies within existing frameworks.

The trajectory, however, is toward more scrutiny. APRA’s 2025–26 Corporate Plan set out targeted AI supervisory engagements with larger entities. The April 2026 letter reported the findings and called for a step-change in AI risk management and governance. The 2026–27 Corporate Plan says cyber and AI risks are intensifying and that entities should expect more frequent and deeper engagement on them.

The message is clear: deploy deepfake detection if you need it — and given the current state of deepfake fraud in insurance, many insurers will — but do so within a framework that APRA can examine and trust.

Preparing for Regulatory Engagement

When APRA engages with an insurer on AI fraud detection, it is likely to want to understand:

Why the insurer chose to deploy (or not deploy) deepfake detection
How the technology was selected and evaluated
What governance surrounds its use in claims decisions
Who is accountable for its performance and outcomes
Whether the insurer can demonstrate fair treatment of policyholders

Insurers should be prepared to walk APRA through the complete lifecycle: from the business case, through vendor selection and model evaluation, to ongoing monitoring and incident management. Documentation is the primary evidence of good governance.

Conclusion

APRA’s expectations for AI fraud detection in insurance are embedded in existing prudential standards — particularly CPS 230, CPS 234 and CPS 220, supported by CPG 235 — and reinforced through increasingly active supervisory engagement. Insurers deploying deepfake detection should treat it as part of a critical operation: governed, monitored, tested and accountable.

The cost of weak governance extends beyond regulatory action. An insurer that deploys AI fraud detection without adequate controls risks both legal exposure for errors and reputational damage that no technology can repair.

For insurers assessing deepfake detection within an APRA-regulated environment, deetech™‘s platform provides analysis outputs, forensic reports and audit records for qualified review. Customers and their advisers remain responsible for mapping those materials to their APRA obligations. Request a demo to see how it fits your claims workflow.

This article is for informational purposes only and does not constitute legal, regulatory, or compliance advice. Insurers should consult qualified legal and compliance professionals for guidance specific to their circumstances and jurisdiction.