O2 Says Its Daisy AI Averted Over 1,000 Scam Calls

O2’s AI “granny” Daisy remains publicly documented as a 2024 scambaiting campaign, not a new consumer feature that subscribers can activate. Virgin Media O2’s 2024 annual report records more than 1,000 scam calls as averted after Daisy began answering fraudsters in November 2024.

The project’s central premise still holds: conversational AI can occupy a scam caller without putting a real person’s credentials or money into the exchange. What the published record does not establish is an ongoing Daisy service placed between O2 customers and their incoming calls—a distinction that matters when assessing both the campaign’s result and its practical relevance.

Daisy was an autonomous scambaiter

O2’s launch account from November 14, 2024 describes software that transcribed a caller’s speech, generated a response through a custom large language model and personality layer, then converted the text into spoken audio. The page identifies YouTube scambaiter Jim Browning as a contributor and states that Daisy held some callers for 40 minutes at a time after operating for several weeks.

The grandmother persona gave the system believable reasons to hesitate, misunderstand instructions and wander into unrelated stories. Daisy discussed family and knitting and supplied fabricated personal or banking information when callers pressed for details.

Those behaviours were functional rather than decorative. In an ordinary conversation, confusion and digressions would be obstacles; in this setting, they extended the interaction while denying the caller anything useful. The scammer encountered a synthetic target built specifically to absorb attention.

What the campaign metric actually establishes

The total indicates that Daisy operated beyond a single staged conversation. It does not, however, show how many financial losses were prevented, how much money was protected or whether fraud attempts across O2’s customer base declined because of the project.

“Averted” is best understood as the company’s campaign measure for calls diverted into unproductive conversations. While a fraudster was occupied by Daisy, that caller could not use the same time to pursue another target. That is a plausible operational benefit, but it is narrower than proving that a specific victim or transaction was saved.

The distinction also prevents Daisy from being confused with call screening. A screening service evaluates or handles calls arriving for customers; Daisy instead acted as the destination for calls already directed into a controlled scambaiting setup. The available campaign material does not describe subscriber controls, enrolment, handset integration or general availability.

Why a creator’s expertise mattered

Jim Browning’s involvement makes the project relevant to the creator economy for a reason beyond audience promotion. His established scambaiting work provided domain knowledge about the rhythms of fraudulent calls, the instructions criminals give and the responses that can keep them engaged.

The collaboration therefore resembled specialist product input more than a conventional endorsement. O2 supplied the conversational system and campaign infrastructure, while Browning contributed knowledge developed through investigative content about scam operations. The resulting character needed both technical fluency and an understanding of how callers react when a supposed victim delays or misunderstands them.

That model has limits. Public-facing scambaiting content can educate viewers and expose tactics, but an automated system can be isolated from real accounts and populated with invented information. An individual answering an unsolicited call does not have the same separation from a genuine phone number, identity or device.

The persona exploited scammers’ own assumptions

Daisy’s apparent age and confusion were part of the mechanism. A caller who regarded an older person as an easy target had a reason to tolerate repetition, mistakes and irrelevant stories instead of ending the conversation quickly.

This creates an uncomfortable reversal: the criminal’s willingness to exploit apparent vulnerability becomes the feature that keeps the criminal occupied. It also demonstrates why vocal realism is not evidence of identity. A convincing voice can represent a synthetic character just as easily as the person a listener imagines.

The lesson extends beyond Daisy without requiring consumers to imitate it. Confidence, urgency, familiarity and apparent vulnerability are features of a conversation, not authentication. Verification must come through a separate, trusted channel.

Consumers should not try to reproduce the experiment

Daisy operated with fabricated details and no genuine customer account behind its voice. A person who prolongs an unsolicited call cannot assume the same protection and may reveal information, respond to a malicious instruction or encourage further contact.

Ofcom’s current scam-call guidance advises recipients not to engage or disclose personal or financial information, to hang up when a call appears suspicious and to forward suspected mobile scam numbers to 7726 free of charge. It also recommends contacting an organisation through a number on its official website or an account statement rather than one supplied by the caller.

Daisy’s significance is consequently narrower than the idea of an AI guardian answering every subscriber’s phone. It demonstrated a controlled use of generative voice technology in which wasting time was the intended outcome, false information was deliberate and no real victim needed to remain on the line.

Also read:

• Anthropic Shifts Stance on Gulf State Investments, CEO Admits Pragmatism Over Principles
• OpenAI Research Finds That Even Its Best Models Give Wrong Answers a Wild Proportion of the Time
• 4 Ways Technology Has Changed How We Make and Receive Phone Calls