Services such as Apate and O2’s Daisy use convincing conversations to occupy fraudsters. The bigger question is how that interaction translates into protection for real people.
A scam caller expects the person answering to be a potential victim. Anti-fraud companies are increasingly putting a conversational AI system on the other end instead.
Australian company Apate says its technology redirects suspected scam calls to bots designed to sustain believable conversations. It also offers text-based engagement and extracts information from those exchanges for organizations investigating fraud.
The approach gives defenders another way to study scams while they are happening. Rather than waiting for someone to report a loss, a controlled conversation can reveal what a fraudster wants the supposed victim to do next.
How Apate’s anti-scam bots work
Apate describes a process involving detection, diversion, conversation and analysis. Its voice technology integrates with telecommunications systems, while its text service engages suspected scammers through messaging.
According to the company, those interactions can produce intelligence such as cryptocurrency wallet addresses, scam scripts and other operational details. Apate says its customers include telecommunications providers, financial institutions and government organizations. These descriptions come from the company and should be understood as product claims rather than independent proof of effectiveness.
There is also evidence of adoption from a banking partner. In June 2025, Commonwealth Bank of Australia announced that it was using near-real-time intelligence gathered by Apate’s AI bots. The bank described thousands of bot profiles engaging scammers through voice calls and text conversations.
O2’s Daisy shows another version of the idea
British telecommunications company Virgin Media O2 developed Daisy, an AI persona presented as a talkative grandmother.
In a published update, O2 said Daisy had answered more than 1,000 scam calls and kept some callers engaged for 40 minutes. The company described fraudsters impersonating trusted organizations, following scripts and applying pressure when the supposed victim hesitated.
Those figures are O2’s account of its own initiative. They demonstrate reported engagement, but they do not establish how many people avoided fraud as a result.
Why intelligence may matter more than call length
TechLayers analysis: A long conversation is an easily understood measure of disruption. Its practical value, however, depends on what happens afterward.
Consider a hypothetical exchange in which a scammer provides a payment destination. That information could help a fraud team investigate an account or connect related complaints. But it still requires validation: a detail supplied during a suspicious conversation is a lead, not a final finding.
The same distinction applies to websites, phone numbers and impersonated brands. Collecting information is one stage; confirming its significance and acting on it are separate tasks.
An effective system therefore needs more than a convincing voice. It needs a reliable route from conversation to investigation.
What would demonstrate that the approach works?
Bot counts and conversation totals describe activity. A stronger assessment would examine outcomes.
Useful questions for banks and telecommunications providers include:
- How often does an interaction produce a verified, previously unknown fraud indicator?
- How quickly does that information reach a team able to act?
- What proportion of flagged activity turns out to be legitimate?
- Can prevented losses be distinguished from the effects of other security measures?
- How are sensitive conversation records protected and reviewed?
These are evaluation questions, not findings about either company.
The distinction becomes especially important when scammers automate their own outreach. A bot keeping another automated system occupied may have a different economic effect from consuming a human operator’s time.
Consumers should leave the conversations to controlled systems
O2 advises the public to leave scam-baiting to specialists rather than trying to imitate Daisy. Its guidance is to end suspicious calls and verify the caller through an official contact number.
A professionally managed decoy operates under different conditions from someone answering on a personal phone. The existence of anti-scam bots should not make an unexpected request for money or personal information feel safer.
AI conversations offer a promising source of fraud intelligence. Their strongest contribution will be demonstrated through verified disruption and better protection—not simply through how long they can keep a caller talking.
