A finance employee joins a video call with the company’s chief financial officer and several colleagues. Everyone looks and sounds exactly right. The CFO asks for a series of urgent, confidential transfers, and the employee follows instructions. Only later does the truth come out: everyone else on that call was an AI-generated deepfake.
That scenario is not hypothetical. It happened to global engineering firm Arup in early 2024, costing the company about $25 million. Deepfakes and AI voice cloning have opened a new chapter in social engineering, one where attackers no longer need to fake an email. They can fake a face, a voice, or an entire meeting.
This guide explains what deepfake scams and voice-cloning attacks are, how they work, real-world examples, warning signs to watch for, and how organizations and security professionals are fighting back.
Social engineering is the practice of manipulating people into giving up information, money, or access. Instead of breaking through technical defenses, attackers exploit human trust, urgency, fear, and helpfulness. Understanding the psychology of hackers helps explain why these tactics work so well.
Classic social engineering includes phishing attacks, pretexting phone calls, and business email compromise (BEC), where criminals impersonate executives or vendors to request payments. What has changed is the toolkit. Generative AI now lets attackers produce convincing fake audio and video in minutes, making old scams far more believable.
A deepfake is synthetic media, usually video, images, or audio, created or altered with artificial intelligence to make someone appear to say or do something they never did. The term combines “deep learning” and “fake.”
Deepfakes are built with deep learning models trained on real recordings of a person. Given enough sample footage, these models can map one person’s face onto another’s body, sync lip movements to new audio, or generate entirely new video of a target speaking. If you want to understand the technology behind these tools, our guide to generative AI vs. predictive AI is a good place to start.
Not every deepfake is malicious. The same technology powers film effects, dubbing, and accessibility tools. The security problem arises when it is used to deceive.
AI voice cloning uses machine learning to replicate a specific person’s voice, including their tone, accent, pacing, and speech patterns. Some modern tools can produce a usable clone from just a few seconds of recorded audio.
That audio is often easy to find. Executives speak on earnings calls, podcasts, webinars, and conference stages. Everyday people post videos to social media. Attackers collect these clips, generate a voice model, and then use it to make phone calls or leave voicemails that sound authentic.
Voice-based phishing, known as vishing, has existed for years. Voice cloning makes it dramatically more convincing because the caller no longer just claims to be the CEO. They sound like the CEO.
Most deepfake and voice cloning scams follow a similar playbook.
Attackers identify who to impersonate and who to target. They study organizational charts, LinkedIn profiles, press releases, and social media to learn names, roles, relationships, and how money or data moves inside the company.
Next, they gather audio and video of the person they plan to impersonate. Public interviews, keynote recordings, and social posts provide raw material for training AI models.
Using widely available AI tools, the attacker creates a cloned voice, a deepfake video, or both. Some attackers use real-time tools that alter their face and voice during a live call.
The attacker creates a believable scenario: a confidential acquisition, an overdue vendor invoice, a locked account, or a family emergency. Urgency and secrecy are almost always part of the story, because they discourage the victim from double-checking.
Finally, the attacker requests something valuable, such as a wire transfer, gift cards, login credentials, a password reset, or access to sensitive systems. In some cases, the deepfake is only the first step, used to gain a foothold that leads to a larger breach or even a ransomware attack.
In January 2024, an employee in the Hong Kong office of engineering firm Arup was invited to a video conference that appeared to include the company’s UK-based CFO and other colleagues. All of them were deepfakes. Convinced the meeting was real, the employee made multiple transfers totaling about HK$200 million, or roughly US$25 million, to five bank accounts. The fraud was discovered only after the employee followed up with company headquarters. (CFO Dive)
The case is a turning point because it showed that seeing a familiar face on a live video call is no longer proof of identity.
In May 2025, the FBI warned that malicious actors had been sending text messages and AI-generated voice messages that impersonated senior U.S. officials. The campaign targeted current and former federal and state officials and their contacts, aiming to build rapport before directing victims to malicious links or requesting information. The FBI advised people not to assume such messages are authentic and to verify the sender independently. (AHA News)
Deepfakes are getting harder to spot, but these red flags still help:
The most important warning sign isn’t technical: any request that could cause real harm if it turns out to be fake deserves independent verification.
No single tool stops deepfake fraud. Strong defense combines process, people, and technology.
The most effective control is out-of-band verification. If a request arrives by phone, video, or email, confirm it using a known, trusted contact method, such as calling back a number from the company directory. Never use contact information provided in the suspicious message.
Require multi-person approval for high-value transfers and changes to banking details. Dual authorization makes it far harder for a single deceived employee to cause major losses.
Some organizations and families agree on private code words or verification questions that a deepfake would not know.
IT help desks should follow strict identity verification steps before resetting passwords or enrolling new authentication devices, regardless of who is calling or how urgent the request sounds.
A zero trust security model assumes no user or request is trustworthy by default. Applied to people and processes, that means verifying identity and authority before acting, every time.
Security awareness training should now include deepfake and voice cloning scenarios, not just phishing emails. Employees who know these attacks exist are far more likely to pause and verify.
Organizations can reduce risk by reviewing how much executive audio and video is publicly available and by being thoughtful about what employees share online.
Deepfake detection tools, liveness checks, and AI-driven anomaly detection can help flag synthetic media and unusual behavior. These tools work best as one layer within a broader security strategy. Learn more about how data analytics, AI, and machine learning support cybersecurity.
Deepfakes are one example of a larger shift. Attackers are using AI to write more convincing phishing messages, automate reconnaissance, and scale fraud. Defenders are using AI to detect threats faster, analyze huge volumes of data, and respond automatically.
That makes AI literacy an essential skill for today’s security professionals. Our article on AI in cybersecurity covers how the field is changing, and our guide to securing AI and machine learning assets explains how organizations protect the AI systems they deploy.
Organizations need professionals who understand both traditional social engineering and the new AI-powered variants. Roles such as security analysts, SOC analysts, fraud investigators, and security awareness specialists are on the front lines. Explore the full range of cybersecurity jobs to see where these skills apply.
These certifications help build the right foundation:
CIAT’s Unlimited Certification Exam Retake Policy covers all CompTIA certifications, including Security+, CySA+, and SecAI+. The policy excludes CISSP and EC-Council CEH.
Deepfakes and voice cloning are changing how attackers operate, and organizations need security professionals who can keep up. CIAT’s cybersecurity programs build skills in threat detection, incident response, and AI security, with industry certifications built into your coursework.
Already working in security? The 5-day CompTIA SecAI+ Bootcamp prepares you to defend against AI-driven threats and secure the AI systems organizations rely on, with live online instruction and hands-on labs. Newer to the field? Start with the CompTIA Security+ Bootcamp.
A deepfake scam uses AI-generated video, images, or audio to impersonate a real person, usually to trick victims into sending money, sharing credentials, or granting system access.
AI voice cloning uses machine learning models trained on recordings of a person’s voice. Once trained, the model can generate new speech that mimics that person’s tone, accent, and speaking style. Some tools need only a few seconds of audio.
Yes. Real-time deepfake tools can alter an attacker’s face and voice during a live call, as attackers did when they impersonated an Arup executive and other employees in a video conference that led to a roughly $25 million loss.
Listen for flat emotion, unnatural pauses, odd phrasing, or audio that sounds slightly too clean. The most reliable test is to hang up and call the person back on a number you already know is legitimate.
Vishing, or voice phishing, is a social engineering attack carried out over the phone. AI voice cloning makes vishing more dangerous because attackers can sound exactly like a trusted person.
Finance teams, executive assistants, IT help desks, HR departments, and anyone who can approve payments or reset access are prime targets. Scammers also target individuals through family emergency scams.
Detection tools help, but they are not perfect, and attackers keep improving. The strongest defense combines detection technology with verification procedures, multi-person approvals, and employee training.
Stop the interaction, do not send money or information, and verify the request through a separate trusted channel. Report the incident to your security team and, if needed, to law enforcement or the FBI’s Internet Crime Complaint Center (IC3).
Creating synthetic media is not automatically illegal, but using deepfakes for fraud, impersonation, harassment, or non-consensual content can violate a growing number of laws. Rules vary by state and country.
CompTIA Security+ covers social engineering fundamentals, CompTIA CySA+ covers threat detection and response, and CompTIA SecAI+ focuses specifically on AI security and AI-driven threats.
401 Mile of Cars Way #100, National City, CA 91950
1717 Louisiana Blvd., NE., Suite 208 Albuquerque, NM, 87110
California Institute of Applied Technology participates in the State Authorization Reciprocity Agreements.
© 2026 California Institute of Applied Technology | info@ciat.edu | (877) 559 - 3621 | Privacy Policy
California Institute of Applied Technology has shared ownership and management of two distinct institutions. California Institute of Applied Technology located in California, and California Institute of Applied Technology located in New Mexico.
GI Bill® is a registered trademark of the U.S. Department of Veterans Affairs (VA). More information about education benefits offered by VA is available at the official U.S. government website at https://www.benefits.va.gov/gibill. CIAT is approved to offer VA benefits. Financial aid is available for those who qualify.
* Students are encouraged to take certification exams while actively enrolled in their Bootcamp, Certificate or Degree program. Unlimited certification exam attempts expire 180 days after program completion. Select exams are not eligible for unlimited retakes - see certification exam policy for details. Industry certifications and/or courses may change at any time to address industry trends or improve student outcomes.
