The Rise of Deepfake Vishing: How AI Voice Cloning Is Changing Social Engineering

/ AI & Emerging Threats, Cybersecurity, Fraud Prevention, Social Engineering

The Rise of Deepfake Vishing: How AI Voice Cloning Is Changing Social Engineering

Based on original reporting by Dan Goodin, Ars Technica (Aug 7, 2025). Read the full article here.

The New Voice of Fraud

Most people have heard of scam calls where the voice on the other end sounds eerily like someone you know — a trusted colleague, a family member, or even your company’s CEO. Thanks to AI-powered voice cloning, attackers can now replicate speech patterns so convincingly that victims believe they’re speaking to the real person.

Cybersecurity experts have been warning about these threats for years. In 2023, the U.S. Cybersecurity and Infrastructure Security Agency (CISA) noted that threats from deepfakes and synthetic media were growing “exponentially.” Google’s Mandiant division confirmed in 2024 that such attacks are now executed with “uncanny precision.”

Anatomy of a Deepfake Vishing Attack

Security firm Group-IB has mapped out the typical workflow of a deepfake voice phishing attack:

  1. Collecting voice samples — Just three seconds of audio can be enough, often pulled from public videos, online meetings, or past calls.
  2. Generating the cloned voice — Using AI-based tools such as Tacotron 2, Vall-E, ElevenLabs, or Resemble AI, attackers type scripts that the system “speaks” in the victim’s voice. Safeguards exist, but Consumer Reports found they can be bypassed with minimal effort.
  3. (Optional) Caller ID spoofing — The attacker makes the number appear to belong to the impersonated person or their organization.
  4. Executing the call — Fraudsters may stick to pre-written scripts or respond in real time for more convincing interaction.
  5. Creating urgency — Classic scenarios include a grandchild in trouble, a CEO requesting a wire transfer, or an IT technician demanding a password reset.
  6. Extracting the payoff — Once money, credentials, or assets are transferred, the loss is usually irreversible.

When the Red Team Wins

To test just how effective this tactic can be, Mandiant ran a red team simulation.

They cloned the voice of a manager inside a company and called employees under that manager’s supervision. Using a real VPN outage as cover, they convinced one employee to override Microsoft Edge and Windows Defender SmartScreen warnings — and successfully executed a malicious payload.

As Ars Technica’s Dan Goodin noted, the exercise revealed “the alarming ease with which AI voice spoofing can facilitate the breach of an organization.”

Defending Against Deepfake Vishing

Stopping deepfake vishing doesn’t require advanced tech, but it does require awareness and discipline:

  • Use a “safe word” system — Pre-arrange a unique code phrase with colleagues or family to verify identity.
  • Hang up and call back — Always return the call using a trusted number, not the one shown.
  • Combine both methods for stronger verification.

The challenge is psychological: attackers thrive on creating urgency and emotional pressure. Staying calm and alert — especially when tired or stressed — is key. Unfortunately, that’s also what makes vishing so persistent and dangerous.

🔒 Stop Deepfake Vishing Before It Starts

AI voice cloning is getting better every day — but so are the defences. With Realm ID and RealSign, you can:

  • Verify that every request really comes from the right person.
  • Secure approvals with cryptographic signatures tied to verified identities.
  • Build trust into your workflows — even when attackers try to mimic your team.

👉 Ready to protect your organisation against deepfake vishing? Get in touch with Realm and see how we can help secure your people, processes, and data.