AI-generated voice phishing: how cloned voices are fooling banks and corporations
What is AI voice phishing
Voice phishing (vishing) is a form of social engineering in which the attacker contacts the victim by phone and tries to obtain money, credentials, or access to systems. In its classic form it relied on a human voice and a convincing story. Artificial intelligence has changed the game: modern tools can clone a voice from a short audio sample and reproduce it with high fidelity. The result is a call that sounds exactly like the caller’s boss, colleague, or family member.
How voice cloning works
A voice clone is built in two steps. First the model learns the acoustic characteristics of a target speaker from a few minutes of audio — a voicemail, a YouTube interview, or a recording of a conference call. Then it generates speech in that voice with any text the attacker chooses. Commercial services offer this capability for a few dollars per minute, and open-source models make it available to anyone with a GPU. The barrier to entry has collapsed.
Scale of the problem
Voice phishing has grown dramatically. According to data cited by cybersecurity analysts, incidents increased by 442 percent year over year in 2024, and the trend continued upward through 2025. The FBI’s Internet Crime Complaint Center recorded over 37,000 tech support scam complaints in 2023 alone, with losses exceeding 924 million dollars. Projections estimate that deepfake-enabled fraud could reach 40 billion dollars globally by 2027. In June 2026, Google’s Mandiant unit described a campaign by the Silent Ransom Group that used voice phishing to impersonate IT support and trick employees into screen-sharing, targeting dozens of law firms and financial companies between January and May of that year.
In August 2026, Bloomberg reported a coordinated wave of AI voice clone attacks against investment firms. Point72 Asset Management, Millennium Management, Two Sigma Investments, and Citadel were among the targets. The attackers mimicked executives’ voices to persuade employees to reveal sensitive information or grant system access. Two Sigma confirmed it stopped the attack, while other firms declined to comment. FINRA notified member firms about the attempted break-ins.
Why it is hard to defend against
Traditional authentication assumes that the person on the other end of the line is who they claim to be. AI voice cloning breaks that assumption. A cloned voice can be combined with a spoofed caller ID, making the call appear to come from the executive’s actual phone number. Multi-factor authentication helps, but a well-executed vishing attack can pressure the victim into approving a login code or revealing a password over the phone. The human element remains the weakest link.
What can be done
Organizations should treat phone calls as an untrusted channel, just like email. Policies that require out-of-band verification — confirming a request through a second, independent channel such as a known email address or an in-person conversation — significantly reduce the risk. Security training that simulates vishing attacks has been shown to yield an average 37 times return on investment. On the technical side, voice biometrics and call authentication protocols such as STIR/SHAKEN provide additional layers, but they are not a complete solution. The fundamental lesson is that trust must be earned through verification, not granted based on a familiar voice.
Oznaczenie treści AI
Ten powstał z udziałem sztucznej inteligencji: research oraz tekst przygotował i zredagował agent AI. Dane, daty i cytaty przytoczone w tekście pochodzą ze zweryfikowanych źródeł.
