Test your cyber skills

Insights/ Artificial Intelligence

Defending against AI-powered deepfake impersonation

2,031 verified deepfake incidents in Q3 2025 alone, $1.1 billion in fraud losses. What AI impersonation detection catches, what it misses, and the evidence-based defences that work.

Mono Training team16 min read
Executive summary

Deepfake technology, synthetic media that uses AI to convincingly alter video, audio, or images, now fuels some of the sharpest social engineering threats organisations face. Criminals clone executive voices and faces to issue urgent requests over video calls, phone, or email, leading to fraud and data loss at scale. AI impersonation detection tools can flag some of this synthetic media, but they catch far from all of it, and attackers adapt faster than detection models retrain. This article looks at the evidence, the regulatory landscape, what detection technology actually catches, and the defences that work when detection fails, centred on people, not just technology.

Data sourcesResemble AI, Deepfake Incident Reports (2025) · Surfshark Research (2025) · Gartner AI Risk Management Survey (2025) · Deloitte Center for Financial Services · ASD Annual Cyber Threat Report 2024 to 25 · South Australian Attorney-General's Department; Senator David Pocock's officeFigures reflect publicly reported research as of October 2026. Findings evolve. Verify current studies before citing externally.

01It has already arrived

The scale of the threat, by the numbers

The deepfake threat is no longer emerging. It has arrived. The data from 2025 is unambiguous about both scale and acceleration.

2,031

verified deepfake incidents in Q3 2025 alone.

Resemble AI
$1.1B

in deepfake-related fraud losses in 2025, roughly three times the 2024 total.

Surfshark Research
62%

of organisations hit by a deepfake attack involving social engineering or automated processes.

Gartner, 2025

In Australia, ASD's Annual Cyber Threat Report 2024 to 25 found the average self-reported cost of cybercrime for large businesses up 219% in a year, to $202,700, with email compromise and business email compromise fraud the most reported crimes against businesses. ASD warns that AI almost certainly lets attackers work at greater scale and speed, and that criminals use it to create fake voices and videos. These threats extend beyond financial loss. They erode trust in digital communication and can manipulate markets or sabotage supply chains through fake executive announcements or vendor impersonation.

02Not hypothetical

Real-world case studies: what has already happened

These are not hypothetical scenarios. The incident below is one of the most thoroughly documented cases of 2024 to 2025, illustrating the tangible impact of deepfake impersonation on an organisation that believed its controls were sufficient.

Arup, $25.6 million, January 2024

A finance employee at the UK engineering and design firm Arup authorised 15 wire transfers, around HK$200 million, after joining a video call populated entirely by AI-generated colleagues, including a deepfake of the CFO. The employee's initial suspicion of the email faded once he saw familiar faces on the call. Hong Kong Police investigated, and Arup confirmed the incident publicly in May 2024. There has been no public report of the money being recovered.

This is not an isolated pattern. A Gartner survey of 302 cybersecurity leaders, run in 2025, found that 62% of organisations had experienced a deepfake attack involving social engineering or exploiting automated processes. Common tactics include impersonation of a trusted contact with an urgent demand for payment or information.

In Australia, individual corporate cases are less publicised, but business email compromise, the fraud deepfakes make more convincing, is already the leading cybercrime reported by Australian businesses to ASD.

03The rules are catching up

The regulatory landscape: Australia and global

Regulatory responses to deepfake misuse are accelerating, but enforcement still lags the threat. Organisations should align with these frameworks now, rather than waiting for mandatory requirements to crystallise.

Australia · StateSouth Australia

Laws in force from 3 November 2025 criminalise creating humiliating, degrading or invasive deepfakes of real people, with penalties up to $20,000 or four years' imprisonment.

Australia · FederalMy Face, My Rights Bill

A private senator's bill introduced by Senator David Pocock on 24 November 2025, not yet law, proposing amendments to the Online Safety Act and Privacy Act requiring consent for using a person's face or voice in AI content, with new powers for the eSafety Commissioner.

Australia · FederalCriminal Code Amendment

The Criminal Code Amendment (Deepfake Sexual Material) Act 2024 already criminalises sharing non-consensual sexually explicit deepfakes, with penalties up to six years' imprisonment.

GlobalEU AI Act & US TAKE IT DOWN Act

The EU AI Act requires AI-generated or manipulated content, including deepfakes, to be disclosed as such. The US TAKE IT DOWN Act (2025) criminalises publishing non-consensual intimate imagery, including AI-generated images, and requires platforms to remove it on request. Organisations with EU exposure or global operations should build these into their compliance strategy now.

Regulatory alignment is a floor, not a ceiling. Organisations that treat compliance as the target are already behind the threat. The evidence calls for a proactive, people-first approach that goes well beyond minimum obligations.

04Where defence actually starts

Building a robust culture: the four pillars

The foundation of deepfake defence lies in people. Technology helps, but deepfakes attack trust and process, so the controls that hold are the ones people follow. Effective defence rests on four interlocking pillars.

01

Interactive simulations

Fake executive calls, glitched deepfake video, voice-cloned vishing. Employees learn to spot lip-sync drift and audio mismatch through live drills, not slides.

02

Monitoring & detection

AI-based anomaly detection, liveness checks that verify real-time human presence, and behavioural biometrics that flag mismatched typing or voice patterns.

03

Human-centric zero trust

No one is trusted by default for unusual requests, even from senior leadership. Code phrases for high-stakes communications, plus phishing-resistant MFA.

04

Incident response, phased

Assess, train, test, review. A dedicated response plan for detection, communication, and legal action, prepared before it's needed, not during.

05The human firewall

What good deepfake awareness training looks like

Generic cybersecurity training does not prepare people for deepfake threats. Programs that actually work share specific traits: deepfake-specific indicators (unnatural facial movement, audio glitches, lip-sync drift), scenario-based simulations across email, calls, and video, executive-focused content for the people disproportionately targeted, clear second-channel verification protocols, and ongoing reinforcement: quarterly, not annual.

Effective defence turns employees into a human firewall, not through fear, but through genuine competence built over time.

That competence only develops through repeated, realistic practice, the same principle that runs through every Mono course.

06Where this is heading

Looking ahead: the deepfake threat, 2025 to 2028

Forecasts from research firms point the same way: deepfake-related threats will keep accelerating. Organisations that build their defences now will be substantially better positioned than those who wait.

2025

Deepfake fraud losses reach roughly $1.1 billion, tripling from 2024. Executive impersonation via voice and video becomes a standard attack vector, and cheap AI tools lower the skill barrier for attackers.

2026

Gartner predicted that by 2026, 30% of enterprises would no longer consider identity verification and authentication solutions reliable in isolation, because of AI-generated deepfakes.

2027

Generative AI fraud losses in the US alone projected to reach $40 billion, up from $12.3 billion in 2023, a 32% compound annual growth rate (Deloitte Center for Financial Services).

2028

Gartner predicts that by 2028, one in four candidate profiles worldwide could be fake, as AI-generated applicants and deepfake interviews reach hiring.

The threat is accelerating. But so is the shield, if you build it now.

·Key takeaways

What this means for your organisation

01The threat is real and accelerating

2,031 verified incidents in Q3 2025 alone, $1.1 billion in losses, 62% of organisations already affected. This is a present risk, not a future one.

02Technology alone is insufficient

Deepfakes exploit human trust, not software vulnerabilities. Detection tools help, but trained, vigilant people are the primary defence.

03Verification protocols are essential

Code phrases, second-channel verification, and zero-trust habits need to become embedded behaviour, not procedures only remembered under pressure.

04Proactive programs outperform reactive ones

Organisations that build people-first simulation programs now will be measurably more resilient than those waiting for a breach to act.