\
Insight
\
\
Insight
\
How agentic AI is scaling the threat of deepfake fraud
Ben Colman
Co-Founder and CEO
Published February 6, 2025 Updated November 5, 2026
Agentic AI scales deepfake fraud by taking the human out of the attack. An autonomous agent can place calls, join meetings, and hold live conversations in a cloned voice or face, so one attacker can run many impersonations at once. Detection on the call or meeting itself flags what scripts and checklists miss.
Deepfake fraud used to take effort. An attacker had to record a voice, write a script, and make each call personally. Agentic AI, meaning AI that can take actions on its own, removes that ceiling. Gartner reported that 62% of organizations experienced at least one deepfake attack in the past 12 months. Autonomous agents make every attempt after that cheaper.
What agentic AI changes about impersonation
Until recently, many deepfake scams used a pre-recorded clip or a single live call run by a person. Agentic AI connects voice cloning, video synthesis, and a language model that can respond to whatever the target says. The result is an impersonator that listens, adapts, and keeps going. Nobody has to sit behind it.
The inputs are minimal. Microsoft Research demonstrated voice cloning from a three-second audio sample. Video has followed into live calls. A 2025 academic paper states that real-time deepfakes are being used for phishing attacks during video meetings.
Not every AI agent is an attacker. Customers, partners, and vendors use AI agents for routine calls too. So the question for a security team isn’t whether an AI is on the line. It’s whether the voice or face is being used to deceive. For the contact center side of that problem, see the agentic AI caller problem.
What deepfake fraud costs as it scales
Each successful impersonation is expensive on its own. Among organizations that experienced deepfake attacks, 55% incurred monetary losses averaging $280,000, according to Ironscales’ 2025 AI Deepfake Threat Report. Agentic AI doesn’t change the cost of one attack. It changes how many attempts an attacker can afford to make. For the wider picture, see the state of AI fraud.
There’s an operational cost too. AI callers that reach a human agent take up the same time as a customer, and that cost lands whether or not the call is fraud. For more on it, read the efficiency cost of agentic AI callers.
Where AI impersonation disrupts business operations
Video meetings
Engineering firm Arup lost $25M when an employee joined a video call with deepfaked versions of the CFO and colleagues and made 15 transfers. An AI agent can now sit in that meeting as a familiar executive. Approval workflows become the attack surface. See how to keep virtual enterprise meetings safe from AI impostors.
Phone calls
AI agents can call help desks and contact centers, move through phone menus, and answer security questions in a cloned voice. Social engineering that once needed a person on every call now runs in parallel.
Executive impersonation
In 2025, an impostor used AI-generated voice and text messages to impersonate Marco Rubio and contact senior foreign and U.S. officials. The same approach can target a CFO, a treasurer, or a board member. See securing financial leaders against AI impersonation.
Why existing controls miss AI impersonators
Most identity controls were built to stop a person pretending to be someone else, not a machine that sounds like them. Challenge questions, caller ID checks, and voice recognition all assume the voice itself is genuine.
That assumption is failing. A 2025 study found that face-to-voice attacks can generate synthetic voices from a single facial image and bypass industry-standard voice-authentication systems.
People don’t catch it either. In a 2025 spearphishing experiment, 66% of participants failed to identify AI-created audio and 43% failed to identify AI-created video as fake.
How detection flags AI impersonation before the decision
Deepfake detection checks the media itself. It analyzes the voice or video for signs of AI generation or manipulation while the call or meeting is still live. The signal arrives while someone can still act, before anyone approves a payment, resets an account, or shares sensitive data.
In meetings, RealMeeting analyzes participant audio and video as the meeting runs. On phone lines, Reality Defender’s AI caller detection flags which incoming calls carry AI-generated voices. Your team decides whether to route a legitimate AI agent, step up verification, or escalate.
Frequently Asked Questions
What security teams ask about agentic AI impersonation
Agentic AI takes the human out of the attack. An autonomous agent can place calls, join meetings, and respond in a cloned voice or face, so one attacker can run many impersonation attempts at once.
No. Customers, partners, and vendors use AI agents for routine calls. Detection flags which calls carry AI-generated voices, so a contact center can route legitimate agents and stop the ones posing as a real person.
Not on its own. Multi-factor authentication confirms a device or a code, not whether the person on a call or in a meeting is real. Pair it with detection on the live voice and video for high-risk decisions.
Before the high-risk action, not after it. Detection runs alongside the live conversation, so the signal reaches the team while someone can still stop a payment, an account reset, or a data request.
Where to start protecting high-risk calls and meetings
Start with the calls and meetings where money or access moves, such as payments, account resets, and executive approvals. Agentic AI turns deepfake fraud from a one-off scam into a volume business. The organizations that hold up check the voice and video behind those decisions, not just the credentials.
Reality Defender was named a Market Shaper, and placed highest on Potential to Execute, in Gartner’s first Emerging Market Quadrant for Deepfake Detection — Startup Vendors, as of June 2026. Get the report
\
Insights