From Detection to Disruption: Inside the AI Shift Redefining Scam Defence
.png)
Table of contents
From Detection to Disruption: Inside the AI Shift Redefining Scam Defence
Executive Summary
- Apate.ai is deployed inside Commonwealth Bank, Australia’s largest bank, engaging scammers at scale
- CBA’s “Pollen” team is running thousands of live conversations simultaneously
- This approach has delivered up to 100 times the scale of previous manual intelligence gathering
- Scam defence is shifting from reactive detection to active disruption
- Real-time intelligence is becoming foundational to modern fraud defence
A system already operating at scale
The Australian Financial Review recently stepped inside Commonwealth Bank’s scam defence centre in Sydney. What it found was a system embedded within live fraud operations and running continuously at scale. A small team now oversees thousands of concurrent scam interactions, handled by AI agents operating in parallel across channels.
These agents engage scammers directly, extract intelligence, and feed it back into fraud systems in near real time.
.png)
From detection to engagement
Most fraud systems are built to detect suspicious activity after it has already begun.
They depend on downstream signals such as transaction anomalies, customer complaints, behavioural flags, or chargeback events. These signals remain critical, but they are reactive by design.
They surface once a scam is underway and money is moving, credentials are compromised, or a customer has already been targeted. This introduces structural latency into the system.
By the time an alert is triggered and verified, the scammer has often rotated phone numbers, mule accounts, domains, or messaging channels. Fraud teams are left responding to fragments of a campaign that may already be evolving.
%20(1).png)
Leveraging Apate’s platform, CBA’s approach has moved earlier in the lifecycle.
Instead of waiting for harm signals, the platform engages scammers during the interaction itself and captures intelligence as it unfolds.
.png)
The technology behind the engagement
The bots engaging scammers inside CBA’s environment are powered by Apate.ai.
These AI agents are designed to convincingly mimic human behaviour across messaging channels. They reflect their deployment demographic, mirroring dialect and behavioural nuance while responding with natural pacing and scepticism as scammers adapt their tactics.
The objective is structured intelligence extraction at scale.
.png)
Since going live last August, the system has conducted more than 272,000 scam interactions, generating payment instructions, account details, and campaign infrastructure signals. That intelligence is reviewed, validated, and fed back into fraud systems to support earlier intervention.
What this scale changes operationally
Before this shift, intelligence gathering was manual. Analysts responded to scam messages directly, made outbound calls, and documented findings one interaction at a time.
Output was constrained by human bandwidth. Coverage slowed outside business hours, and scaling required additional headcount.
.png)
Now, a small team oversees more than 2,000 live scam conversations running simultaneously, increasing intelligence output by up to 100 times compared to manual engagement. Overnight, the system surfaces security alerts, payment instructions, and infrastructure details ready for review the next morning.
Rather than responding to isolated incidents, teams begin with a structured view of patterns emerging across thousands of interactions.
.png)
Why this model matters now
Scam operations no longer resemble isolated incidents. They operate more like coordinated enterprises, testing scripts, rotating infrastructure, and shifting channels as soon as pressure appears.
Defensive systems, by contrast, are still largely structured around detection. Transaction monitoring, behavioural analysis, and customer reporting remain essential, but they activate once an attempt has already progressed. By the time an alert is investigated, the infrastructure behind it may already have changed. That timing gap reflects structural design rather than a lack of effort. Criminal networks iterate continuously, while many defensive controls are built to respond to outcomes.
Engagement-based intelligence shifts that balance.
When an organisation interacts directly with scam infrastructure, it can see campaigns before they fully mature. Payment instructions surface earlier. Mule accounts can be identified before widespread reuse. Patterns emerge across conversations that would otherwise remain fragmented.
This does not replace traditional detection systems. It strengthens them by moving visibility further upstream, where disruption remains possible.
.png)
The acceleration on both sides
Artificial intelligence is reshaping both defence and offence.
The same advances that allow institutions to analyse data faster and automate complex workflows are available to criminal networks. Tools that generate convincing text, voice, and video are becoming cheaper and more accessible, increasing both scale and sophistication.
Fraud teams are no longer improving controls within a stable environment. They are adapting within a landscape that evolves more quickly each year, where campaigns can be launched, tested, and refined in compressed timeframes.
In this context, speed becomes a defensive capability in its own right.
The ability to generate intelligence while a scam is still unfolding materially changes how quickly an organisation can respond. It reduces reliance on lagging indicators and increases the proportion of decisions informed by live signals.
As automation expands on both sides, the question is not whether AI will shape scam defence, but whether institutions can use it to move earlier and act with greater certainty.
.png)
A shift in operating model
What is happening inside CBA’s scam defence centre represents a paradigm change in how scam risk is approached.
Instead of relying solely on downstream detection, the model incorporates direct engagement with adversaries as a source of structured intelligence. That intelligence feeds existing fraud systems, strengthens decision-making, and improves response speed without replacing established controls.
If scams are increasingly automated and coordinated, defensive strategies must evolve in parallel. Visibility needs to move earlier in the lifecycle, and intelligence must be generated rather than passively received.
If you are rethinking how your organisation surfaces intelligence and disrupts scam infrastructure earlier in the lifecycle, this model is already in production. It is worth understanding how it fits within your existing fraud stack.
.png)
Let’s work together.
We work closely with each client to understand their unique requirements and provide a solution that fits. Reach out for a personalised consultation and to explore how our technology can transform your scam prevention and intelligence strategy.
