Daisy, the AI ​​Grandma Who Turns Phone Scams on Their Heads

In the UK, they've decided to take on phone scammers with a simple yet brilliant idea: a virtual grandmother who never stops talking. The star is 'Daisy', an AI created by Virgin Media O2 to ensnare scammers in endless calls and, incidentally, prevent them from spending that time deceiving real people.

The premise is straightforward and effective: if a scammer wastes 40 minutes with Daisy, those are 40 minutes in which they won't manage to steal anyone's data or money. The company claims to have trained Daisy with real content from scammers and scambaiting techniques so that her way of speaking and reacting is almost indistinguishable from that of a kind, absent-minded, and chatty older person.

What is Daisy and why does it exist?

Daisy is a combination of several artificial intelligence models that cooperate to listen, understand, and respond over the phone in real time. Virgin Media O2 presents her as a 'new member' of the fraud prevention team , with the ironic title of 'director of fraudster relations'.

Her role is to do what no one wants to do: endure the con artists' chatter and thwart their attempts. While entertaining them with family stories, sewing anecdotes, or distractions about her fictional pet , Daisy drops false clues and fabricated facts that never lead anywhere.

The idea has a clear social component. O2 detects that 71% of the British population would like to stand up to scammers , but without sacrificing their time; that's where Daisy fits in, who literally 'has all the time in the world' and doesn't get desperate.

How it works and what technologies it uses

To sound like a real grandmother, a single model isn't enough: Daisy relies on a selection of generative models and a trained LLM to fine-tune her personality and knowledge base . Her voice was also modeled after the grandmother of a team member, giving it a believable and approachable timbre.

The public image of the character for the campaign was generated using a custom-trained broadcast model. The entire project was developed in collaboration with Faith, VCCP's AI creative agency , as part of the 'Swerve the Scammers' initiative, which aims to raise awareness about telephone fraud.

The key is naturalness. Daisy recognizes typical conversation patterns of scammers , maintains an affectionate and distracted tone, and responds with pauses, trivial questions, or micro-stories that lengthen the call without providing anything useful to the criminal.

Examples of calls and tactics that dismantle

When a con artist tries to sell her a supposed electricity bill refund, a miraculous job offer, or a fabricated technical problem, Daisy changes the subject without seeming rude. She might ask for time to find her glasses, talk about a scone recipe, or discuss a knitting pattern while the con artist loses their train of thought and their patience.

If asked to download an app or open a link, it feigns technical uncertainty with ease. The AI ​​even provides fake bank details or fabricated information to reinforce the illusion that it's the perfect victim , but it never exposes real information or steps that could put users at risk.

In internal tests, Virgin Media O2 indicates that Daisy has managed to keep numerous scammers on the phone for approximately 40 minutes at a time. This is enough time to prevent them from attempting another call to a vulnerable person and, at the same time, to study the most common tactics they are using.

Impact of fraud: painful figures

The problem is significant. BDO estimates that the total value of fraud in the UK reached £2.300 billion in 2023 , double the previous year's figure. Fraudulent calls continue to rise and disproportionately affect those with less technological familiarity.

The University of Portsmouth found that two-thirds of those surveyed over the age of 75 had experienced at least one attempted fraud in the past six months. Older people are a prime target for scammers , and this is precisely the bias Daisy exploits to deceive them in return.

Furthermore, according to O2 data, 22% of Britons face attempted fraud weekly, and 67% live in fear of becoming a victim. Given this situation, any tool that reduces the time and effectiveness of criminals contributes to overall protection.

Training with real material and expert collaboration

To prevent Daisy from sounding robotic, she had to be trained with authentic material. Virgin Media O2 states that it used real content from scammers and relied on Jim Browning , the well-known creator who investigates fraudulent call centers and popularized scambaiting on YouTube.

With Browning, the company fed Daisy's phone line with numbers that scammers commonly dial. Decoy phones were also placed in locations where scammers scout for potential victims , so the AI ​​would receive real calls under real-world conditions.

The result: Daisy has been operational for weeks interacting with fraudsters in real time without direct human intervention , confirming that her behavior holds up outside the lab and against multiple accents, rhythms, and fraud scripts.

Campaign objectives and actual reach

Virgin Media O2 has been clear: the project has a strong educational component. The goal is not to deploy millions of 'Daisy' phones simultaneously, but to raise awareness of criminal tactics and encourage the public to be alert to any suspicious calls.

This educational approach relies on audiovisual pieces featuring well-known personalities. Influencer Amy Hart, a victim of bank fraud, collaborates on the campaign to give voice to the real-life experience and explain how to identify warning signs.

Meanwhile, O2's fraud team emphasizes: it's best not to engage in scambaiting on your own. They recommend reporting suspicious calls and messages to the free number 7726 so the operator can investigate and block them.

How calls are routed and why it's not on your phone

A common question is whether Daisy can be installed on a personal phone. It can't: the AI ​​operates from specific lines listed by criminals , so when they're looking for victims, they end up encountering it.

Another entry point is through complaints. By forwarding communications to 7726, O2 identifies patterns and sources , and can redirect high-risk calls to Daisy to weed scammers out and away from legitimate customers.

This design minimizes risk to users and maximizes the AI's effectiveness as a time-trapping tool. The company itself asks users to leave the "revenge" to Daisy and simply report incidents to stop the fraud at its source.

More measures against spam and impersonation

Daisy is not alone. O2 has launched AI-powered anti-spam tools and free caller ID services that help filter out suspicious numbers before they reach the recipient.

The telecom operator claims that it blocked 89 million fraudulent SMS messages and stopped over £250 million in suspicious transactions last year. These figures give an idea of ​​the scale of the problem and the effectiveness of the filters when fed with reports to 7726.

Finally, the company has demanded a stronger response from the British government. It proposes appointing a dedicated minister to combat fraud and creating a single, well-funded police force to centralize all investigations.

Creative details that make a difference

Daisy's success is no accident: her character was meticulously crafted. Her voice was modeled after that of a team member's grandmother, lending it genuine texture, rhythm, and expressiveness ; and her visual representation for the campaign was developed using a custom-trained broadcast model.

This technically creative work translates into conversational nuances that fool anyone. The AI ​​strings together short, believable stories about family, household chores, or hobbies , and displays credible technological doubts that fit the stereotype scammers expect.

The goal is to exploit criminal prejudices and turn them on their head. If the scammer thinks they've found the perfect victim and gets overconfident, the AI ​​further ensnares them with pointless requests, comical misunderstandings, and calculated pauses.

Measured results and continuous learning

The average of 40 minutes isn't a ceiling, but rather a demonstration of capability. Each call serves to reinforce pattern recognition, refine the script, and improve Daisy's resilience to emerging techniques.

In addition to direct blocking, there's intelligence value. AI allows for the analysis of social engineering scripts, accents, rhythms, and counterarguments that work —data that can be applied to other layers of the operator's defense.

This knowledge helps to better protect real users, who are at the heart of the equation. The more we learn about criminals, the easier it is to prevent their campaigns from happening.

Ecosystem context: expanding threats

The rise in fraud is not limited to the telephone. Cases such as the attempted deepfake scam targeting the CEO of WPP or the increasing regulation of weak passwords on connected devices demonstrate the breadth of the problem.

Specialized technology companies indicate that the use of AI in defense is expanding to banking, insurance, and the public sector. It is being used to detect unauthorized transactions, review dubious claims, and identify tax irregularities —areas where fraud is also trying to gain a foothold.

All of this leads to a clear conclusion: there's no silver bullet, but there are combined strategies that reduce the impact . Within that range, Daisy occupies a unique position: taking time away from scammers and creating more room to protect customers.

Advantages, limits and ethical issues

The main advantage is clear: turning the AI's infinite patience into a resource that scammers can't reclaim . Every minute spent on Daisy is a minute that doesn't reach a potential victim.

Among the limitations, the operator himself points out that he does not intend to universalize this approach. It is a tool for raising awareness and containing crime, not a substitute for self-protection or coordinated police action.

From an ethical standpoint, the project was designed to avoid encouraging individuals to play with fire. The guideline is to channel all attempts to 7726 and avoid personal risks , keeping the confrontation with the scammers in a controlled environment.

The role of citizens: what you can do

The first rule is simple: be wary of unsolicited calls asking for personal information or claiming payment urgency. At the slightest suspicion, hang up and contact them through official channels . Your bank, government agency, or mobile provider will never unexpectedly demand passwords or codes over the phone.

Second, take advantage of your carrier's tools. If you use O2, forward suspicious messages and numbers to 7726 ; these reports are the fuel that allows for large-scale blocking and feeds systems like Daisy.

Third, share this information with those around you. Older relatives and those less tech-savvy are common targets ; a timely conversation about signs of fraud can save you trouble and money.

A very skilled virtual grandmother

Between jokes about sewing and innocent questions about technology, Daisy employs a serious tactic: turning social engineering on its head and exploiting the con artist's biases against him . That's where its effectiveness lies, beyond the lighthearted banter.

That an AI can sustain a believable 40-minute conversation without losing its composure speaks volumes about the state of the art. The coordination between generative models, a credible synthetic voice, and a well-tuned LLM personality sets a high bar for future conversational defenses.

With the support of a powerful creative campaign and the reinforcement of network tools, Daisy fits like a piece of a larger puzzle. Automatic blocking, caller ID, pattern analysis, and institutional coordination complete the framework that truly reduces fraud.

After learning how it works, it's clear that it's not meant to be a complete solution or to live on your phone. It's primarily a vivid reminder that the person on the other end of the line might not be who they claim to be , and that today the best defense combines technology, patience, and common sense.