The Scams Prevention Framework Act 2025 obliges designated businesses, starting with banks, telecommunications providers and certain digital platforms, to govern, prevent, detect, disrupt, respond to and report scams. AI does real work on detection and disruption, scoring payments and messages in close to real time, but it must be explainable, reconciled with the Privacy Act, and owned by a person.
The Act received Royal Assent on 20 February 2025. It creates an enforceable, principles-based regime with regulator oversight, intelligence reporting, external dispute resolution through the Australian Financial Complaints Authority, and civil penalties that reach tens of millions of dollars per contravention. Obligations bite as each sector is designated and its rules and code commence. Banking, telecommunications and key digital platforms were designated in May 2026, with obligations expected to apply from 31 March 2027 (Treasury, 2026), so confirm current commencement dates for your sector with Treasury and the ACCC.
The compliance question is no longer whether you have an anti-scam page. It is whether you can show, on the evidence, that your systems detected and disrupted scam activity across your customer base. This guide is for risk, financial crime and technology leaders at banks, mutuals, telcos and platforms.
Key takeaways
- The SPF rests on six principles: govern, prevent, detect, disrupt, respond and report.
- Compensation under the respond principle turns on whether you met the other five, so detect and disrupt evidence decides liability.
- Behavioural models beat static rules for detection; disruption needs inline scoring before settlement or delivery.
- AI scam controls still answer to the Privacy Act and Australian Consumer Law, and must be explainable.
- False positives are the biggest practical failure: tuning is continuous, owned work.
Why does the Scams Prevention Framework exist?
The National Anti-Scam Centre's Targeting Scams report for 2024 recorded combined reported scam losses of $2.03 billion in 2024, drawn from Scamwatch, ReportCyber, the Australian Financial Crimes Exchange, IDCARE and ASIC. That was down 25.9 per cent on 2023, which the ACCC attributes to coordinated effort across government, law enforcement and industry. The report recorded 494,732 scam reports.
Two things follow. Falling losses show coordinated intervention works and is worth mandating. And two billion dollars of reported loss, with under-reporting widely acknowledged, is enough harm for government to legislate sector duties instead of relying on voluntary codes.
What are the six SPF principles?
Six overarching principles apply to every regulated entity, supported by sector codes for banking, telecommunications and digital platforms.
- GovernSenior-accountable anti-scam governance and strategy
- PreventStop scams reaching customers in the first place
- DetectIdentify scam activity while it is happening
- DisruptIntervene mid-flight to stop loss
- RespondHandle reports and remediate where duties are breached
- ReportShare intelligence with regulators and consumers
- Govern: a senior owner and a documented strategy.
- Prevent and detect: intelligence and monitoring duties.
- Disrupt: acting on a suspicious transaction or message before money or data leaves.
- Respond: complaints handling and possible compensation where obligations were not met.
- Report: sharing actionable scam intelligence with the ACCC, which can alert other entities.
Note where the legal weight sits. Compensation under respond depends on whether you met the other five, so your detect and disrupt evidence determines liability when a customer is defrauded.
Where does AI do real work under the SPF?
AI is a set of capabilities mapped to specific principles, not one product bolted on.
| Manual or Rules-Only | AI-Augmented | Impact | |
|---|---|---|---|
| Transaction monitoring | Static thresholds, high false positives | Behavioural anomaly detection per customer | Detect |
| Scam message analysis | Keyword blocklists, easily evaded | Pattern and intent classification at scale | Detect |
| Real-time intervention | Post-event review, money gone | Inline risk scoring before settlement | Disrupt |
| Intelligence sharing | Manual report compilation | Automated structuring of actionable signals | Report |
| Governance evidence | Periodic spreadsheets | Continuous control logging | Govern |
Banks. Static rules such as flagging every transfer above a fixed amount drown analysts in false positives and miss scams hidden in ordinary amounts. Behavioural models that learn a customer's normal pattern and score deviation against known typologies, such as a first payment to a newly added payee after a run of inbound calls, separate risk from noise far better. It is the same discipline as AI fraud detection, now in a mandatory context.
Telcos. Detection focuses on call and message traffic. Classifiers that recognise scam SMS signatures or number-spoofing call patterns operate at volumes no team can match.
Digital platforms. A sector obligation is verifying advertisers hold appropriate licences before advertising high-risk products such as financial services. AI helps with document checks and behavioural anomalies; the verification decision stays a controlled, auditable step.
Disruption is where detection becomes prevented loss. A risk score only helps if it is acted on inline, before a payment settles or a message is delivered. That needs low-latency scoring in the transaction or messaging pipeline, not a nightly batch job.
What does a detect-to-disrupt loop look like?
For example, consider a hypothetical mutual bank with several hundred thousand customers and a lean financial crime team.
- InitiateCustomer initiates an unusual payment
- ScoreModel scores risk against scam typologies
- HoldHigh-risk payment paused for verification
- ConfirmCustomer contacted with targeted warning
- ReportConfirmed scam signal shared to ACCC
Intercepting a single high-value scam payment can offset a meaningful share of a model's running cost, while doing nothing demonstrable becomes a source of penalty and compensation exposure. The point is the shape of the loop, not a promised return.
- Customer funds intercepted before settlementDirect loss avoided
- Compensation exposure under the respond principleLiability reduced
- Civil penalty exposure for inadequate systemsRisk reduced
- Regulator and ombudsman confidenceEvidence on record
What risks does AI introduce for scam prevention?
Deploying AI to satisfy one law does not exempt you from others.
Privacy. Behavioural detection profiles customer activity, which is personal information under the Privacy Act 1988. Reconcile models with the Australian Privacy Principles, especially purpose of collection and data shared under the report principle. Privacy reforms adding transparency for automated decision-making raise the bar; see automated decisions and the Privacy Act.
Fairness and false positives. Wrongly holding legitimate payments, or flagging some customer groups disproportionately, creates its own harm and possible Australian Consumer Law exposure. An automated scam warning is a representation that must not mislead. See consumer law and AI.
Explainability. A regulator will expect you to explain why a model held or cleared a payment. Opaque scoring with no audit trail fails. A person owns the decision and the system records why, as in our AI agent governance framework.
Before you deploy, can you answer yes to each?
- Can you explain any individual hold or release?Explainability: required for the govern principle
- Is the profiling reconciled with the APPs?Privacy: required under the Privacy Act 1988
- Do customer-facing messages avoid misleading claims?ACL: required under the Australian Consumer Law
- Does a human own the final intervention call?Governance: required for accountability
- Is every decision logged for the regulator?Evidence: required for reporting and disputes
Should you build, buy or partner for scam detection?
Few regulated entities should build a scam detection stack entirely in-house. The realistic options are a specialist vendor platform, an internal build on cloud machine learning services, or a hybrid where a partner speeds the build while your team keeps governance. The decision turns on data sensitivity, in-house capability and how much model logic you must control and explain.
Data location matters. Models train on highly sensitive financial and behavioural data. Know what leaves your environment and never send regulated customer data to a tool whose data handling you have not verified. Our data sovereignty guide covers the options.
How do you tune false positives in an AI scam control?
The biggest practical failure is stopping legitimate activity. Too many holds, or warnings so frequent they become noise, erode trust, drive complaints and train customers to click through every alert, defeating disruption.
A defensible program tracks how often holds turn out to be scams versus legitimate, segments that by customer group to check for unfair impact, and feeds confirmed outcomes back into the model. When a customer disputes a loss, a well-tuned, well-evidenced system is your defence. A blunt one you cannot explain shows monitoring without control.
How should a regulated entity sequence the build?
- Phase 1Govern and mapAppoint the senior owner, document the strategy, map your data and current controls against the six principles
- Phase 2Detect baselineDeploy behavioural monitoring on the highest-risk channels, tuned to reduce false positives
- Phase 3Disrupt inlineIntegrate real-time scoring into the transaction or messaging pipeline with a human-in-the-loop hold
- Phase 4Report and improveAutomate structured intelligence sharing and feed confirmed outcomes back into the models
Govern first, because it sets accountability and priorities. Detect before disrupt, because intervening on a poorly tuned model harms customers. Reporting last, because shared intelligence is only as good as the detection behind it.
Not yet designated? Super funds, crypto exchanges and adjacent businesses should note the framework is built to expand. Treating the six principles as a design standard now shortens your runway later. See also AI compliance under APRA and ASIC.
How Solve8 can help
Our consultants and engineers help regulated entities map controls to the six principles, connect scoring into payment and messaging systems, and build the audit trail regulators expect, on your infrastructure in Australia with a person owning each intervention. See our financial services work and system integration service, or contact us to scope the first phase.
