For years, the standard advice for a professional voicemail greeting was to record it yourself, state your name and company clearly, and sound approachable. That advice made sense when the worst thing a caller could do with your greeting was hang up. It no longer makes sense today, and I want to walk through why, because the reasoning matters more than the rule.
Modern voice cloning platforms require remarkably little source material to produce a usable result. Transaction Network Services and others have documented that roughly three seconds of clean audio is enough to produce a convincing synthetic copy of a person's voice. Three seconds. Your voicemail greeting, the one that says "Hi, you've reached Jeff Lauria at iCorps Technologies, please leave a message," runs longer than that and hands an attacker three things in a single phone call they never had to answer: a clean voice sample, your full name, and your employer. They now know who you are, where you work, and what you sound like, and they collected all of it without speaking to a human being. From there the workflow is unfortunately simple. A quick pass through LinkedIn fills in your title and your reporting relationships, and the attacker has everything required to place a call to your controller, your service desk, or your family that sounds exactly like you.
AI Voice Cloning Attacks Are Already Happening
The aggregate numbers tell part of the story. CrowdStrike's 2025 Global Threat Report documented a 442 percent surge in voice phishing incidents between the first and second half of 2024, and in May 2025 the FBI issued a public alert warning that threat actors were using AI generated voice messages to impersonate senior U.S. officials, with the explicit goal of building rapport before pivoting to credential theft. Industry projections put deepfake enabled fraud losses in the range of 40 billion dollars by 2027. Those are large numbers, and large numbers have a way of feeling abstract, so let me make it concrete with incidents that actually happened.
In July 2024, an executive at Ferrari received WhatsApp messages and then a live call from someone who sounded precisely like CEO Benedetto Vigna, regional accent included, pressing for help with a confidential transaction. The executive asked a personal question the caller could not answer, and the attack collapsed on the spot. Earlier that same year, a LastPass employee received calls, texts, and voicemails from a cloned copy of their CEO's voice over WhatsApp; the employee recognized the urgency and the odd channel as red flags and reported it rather than responding. And in the case that should keep every CFO up at night, an employee at the engineering firm Arup wired 25.6 million dollars after a video conference in which every other participant, including the CFO, was synthetic.
Notice the pattern in the two attacks that failed, because technology did not stop either of them. A skeptical human being who paused and verified stopped them. That observation should shape how we approach defenses, because the controls that matter here are behavioral, inexpensive, and available to every organization today.
How AI Voice Cloning Changes Voicemail Security
The first change costs nothing and takes five minutes to implement. Stop using your own voice for your voicemail greeting. Every business phone system and every mobile carrier offers a system generated greeting, the flat automated voice that reads back the number. Use it. I recognize this feels impersonal, and for years we trained professionals to believe a recorded greeting in their own voice signaled attentiveness and polish. The threat model has changed, and the etiquette needs to change with it. A synthetic greeting gives an attacker nothing to sample. Legitimate callers, the people who actually know you, will leave a message anyway, because they dialed your number on purpose.
The second change follows the same operational logic: do not identify yourself in the greeting. A greeting that announces your name and company to every anonymous caller is performing reconnaissance on your behalf and handing over the results for free. The system voice reading back the phone number confirms the caller reached the right place, which is the only job a greeting actually has. Anyone who needs more confirmation than that can state their business in the message and wait for a callback. I would extend this to department mailboxes as well. A greeting that recites names, titles, and direct extensions is an org chart delivered on request, and attackers building a pretext for a call to your help desk are grateful for it.
How to Reduce Voice Phishing and Vishing Risk
The third change concerns how you answer inbound calls, and this one requires breaking a habit most of us have carried across our entire careers. When a call comes in from a number you do not recognize, do not answer with your name. Answer with a plain "hello" and nothing more, then let the caller carry the conversation. There are two reasons for this discipline, and both of them matter operationally. The first is the same sampling problem: an attacker who calls you and gets "This is Jeff" has captured a voice sample and confirmed the number belongs to you, and modern cloning tools are effective even with the few seconds of audio that a brief answered call provides. The second reason is that identity confirmation is itself the objective of many of these calls. Robocall operations dial thousands of numbers to sort the live, confirmed lines from the dead ones, and the confirmed list becomes the target list for the vishing campaign that follows.
Pause before you speak. Say hello, and if the line sits silent for a beat, say nothing more and hang up. That silence is frequently an autodialer deciding whether to connect you to an operator or a recording engine. Resist the reflex to fill it. If the caller claims to be your bank, a vendor, or a colleague in distress, disconnect and dial back on a number you already possess from a statement, a contract, or your own directory. No legitimate institution objects to a callback. The ones that object are telling you something.
How Businesses Can Defend Against AI Voice Cloning Scams
Individually these habits sound small, and that is exactly why they work. They remove the raw material the attack depends on, and they cost nothing to implement. But they belong inside a broader posture, because voice cloning is not an isolated trick; it is a delivery mechanism for the same business email compromise playbook we have fought for a decade, now with a voice on the wire that sounds like your CEO.
So establish verification protocols that dinons, share credentials, or grant access gets verified through a second, independent channel, full stop, no matter how senior the requester sounds or how urgent the deadline. Ferrari's executive defeated a sophisticated attack with a single personal question; a standing challenge phrase or a mandatory callback rule accomplishes the same thing without requiring anyone to improvise under pressure. Audit the audio your organization publishes, because webinar recordings, podcast appearances, and conference talks are all sampling material, and while you cannot pull your CEO off the speaking circuit, you can assume their voice is already cloneable and design your controls accordingly. And train for it deliberately, because your finance team and your service desk should hear about these incidents before the attacker calls, not after. The FCC banned AI generated robocalls back in February 2024, and regulators will keep moving, but no enforcement action will arrive fast enough to protect the employee holding a live call from a synthetic voice this afternoon.
The uncomfortable truth is that voice, which humans have trusted as proof of identity for as long as we have had telephones, no longer proves anything. That trust took a century to build and about three seconds of audio to break. The organizations that internalize this early, that treat every voice on an inbound call as unverified until proven otherwise, will weather the next eighteen months considerably better than the ones still assuring themselves they would recognize a fabrication. The Arup employee thought so too, and that call cost 25.6 million dollars.
Update the greeting. Answer with hello. Verify before you move money. None of it requires a budget line, and all of it works.
Concerned about how AI-enabled social engineering could affect your business? Meet with the iCorps team to assess your security posture, identify gaps in your people and processes, and build a practical plan to reduce risk.