AI has made impersonation cheap. The FBI received more than 22,000 complaints with an AI link in 2025, reporting $893 million in losses, and IBM found that deepfake impersonation made up 45% of AI-enabled attacks in its 2026 breach study. One engineering firm lost $25 million after a video call with a faked finance chief. This guide sets out the numbers and the verification habits that defeat these scams.
The numbers at a glance
Data on AI fraud is still young, and each source measures something different. The FBI counts complaints where victims or investigators noted an AI element. IBM measures how often AI was involved in breaches it studied. Deloitte offers a forecast. Read together, they show AI fraud growing from a novelty to a routine part of scams aimed at businesses.
The FBI figures are best read as a floor, not a ceiling, because many victims cannot tell whether a voice or video was synthetic, and many businesses never report fraud at all. Expect the true figures to be noticeably higher than any count shown here.
| Measure | Figure | Source |
|---|---|---|
| Complaints with an AI link, 2025 | 22,000+ | FBI IC3 |
| Losses in those complaints | $893 million | FBI IC3 |
| AI-assisted BEC losses, 2025 | $30 million+ | FBI IC3 |
| AI-enabled attacks that used deepfake impersonation | 45% | IBM 2026 |
| Average cost of AI-driven breach | More than $6 million | IBM 2026 |
| Forecast US gen AI fraud losses, 2027 | $40 billion | Deloitte |
How deepfake fraud works against a company
The best known case is Arup. In early 2024 a staff member in its Hong Kong office received a message that appeared to come from the UK-based chief financial officer about a confidential transaction. The employee then joined a video call where the CFO and other colleagues were AI-generated, and afterwards sent about $25 million (HK$200 million) to five Hong Kong bank accounts. The fraud surfaced only when the employee checked with headquarters.
The FBI has described a related pattern. From April 2025, attackers sent text and AI-generated voice messages posing as senior US officials to build rapport, then moved targets to another messaging platform to steal account access. The common thread is pressure plus a familiar face or voice, used to skip the normal checks.
What breach data says about AI attacks
IBM's 2026 study found that more than one in four organisations hit by a malicious attack reported AI involvement, a 56% increase on the year. AI-related breaches made up 21% of all breaches, up from 13%. Deepfake impersonation accounted for 45% of AI-enabled attacks, AI-enabled malware for 19%, and AI added about $1 million to the average breach cost.
In India, IBM found 26% of malicious breaches were AI-generated. Verizon's 2026 DBIR adds that phone and text phishing succeeded about 40% more often than email in simulations, and APWG recorded vishing up 20% and smishing up 40% between Q1 and Q2 2026. Voice is where AI impersonation does the most damage.
Where the losses are heading
Deloitte's Center for Financial Services estimated in 2024 that fraud losses enabled by generative AI in the US could grow from $12.3 billion in 2023 to $40 billion by 2027, a compound annual growth rate of 32%. Under an aggressive adoption scenario it put generative AI email fraud losses alone at about $11.5 billion by 2027.
Deloitte also noted that scam software is sold on the dark web for as little as $20. The FBI's 2025 data shows the shift already under way in consumer fraud: more than $19 million in romance scam losses and over $5 million in distress scams, such as voice-cloned calls from a supposed relative, had a likely AI connection.
India's rules on synthetic content
India amended its IT intermediary rules to cover synthetically generated information. The Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 were notified on 10 February 2026 and took effect on 20 February 2026. They require platforms that let users create or share AI-generated audio and video to label it prominently and embed metadata that identifies its source.
The rules also cut the time to act on certain lawful takedown orders to 3 hours from 36 hours, and reduce grievance response periods to 7 days from 15. For a business, this gives a route to get a deepfake of an executive or brand removed quickly, but only if someone spots it and files promptly.
| IT Amendment Rules 2026 | Detail |
|---|---|
| Notified | 10 February 2026 |
| In force | 20 February 2026 |
| Takedown on certain lawful orders | 3 hours (was 36 hours) |
| Grievance response period | 7 days (was 15 days) |
| Labelling | Prominent label and embedded metadata on synthetic content |
Defences: build verification into payments
Deepfakes beat recognition, so do not rely on recognition. Make the rule that no payment, bank detail change or credential reset is approved on the strength of a call, video or message alone, however senior the person appears. Confirm by calling back on a number already in your records, and require a second approver for amounts above a set limit.
The FBI recommends verifying any new contact details through a previously confirmed channel, never sharing one-time codes, and agreeing a secret word or phrase to confirm identity. For a finance team, that can be a code phrase that changes each quarter. XDQ Labs builds such checks into the approval flows of the systems it develops.
- Callback on a known number for every payment or bank change
- Dual approval above a fixed amount
- A rotating code phrase for urgent executive requests
- No MFA codes or passwords shared over any channel
- Help desk identity checks before resets
- A named person to call when something feels off
Training staff to spot synthetic media
Detection by eye is unreliable, but it still helps to know the signs. The FBI lists distorted hands or faces, inconsistent shadows, lag on calls, unnatural movement, and odd tone or word choice as clues. Urgency, secrecy and a request to move to a new app are stronger signals than any visual glitch.
Run short exercises that include a fake voice note or a surprise video call request, and praise people who pause to verify. Publish official channels for executives and the brand, monitor for impersonation, and decide in advance who reports a deepfake to the platform and, in India, to the cyber crime portal.
Frequently asked questions
How common is deepfake fraud against businesses?
IBM's 2026 breach study found more than one in four organisations hit by a malicious attack reported AI involvement, and deepfake impersonation made up 45% of those AI-enabled attacks. The FBI received more than 22,000 complaints with an AI link in 2025, reporting $893 million in losses.
What happened in the Arup deepfake scam?
In early 2024 an employee in Arup's Hong Kong office joined a video call in which the UK chief financial officer and other colleagues were AI-generated. Believing the call was real, the employee transferred about $25 million (HK$200 million) to five bank accounts. The fraud was found when the employee checked with headquarters.
How can a company protect itself from voice cloning scams?
Never approve payments or resets on a voice or video request alone. Call back on a number already on file, require two approvers for larger amounts, and use a code phrase for urgent requests. The FBI also advises never sharing one-time codes and verifying new contact details through a known channel.
What do India's 2026 IT rules say about deepfakes?
The IT Amendment Rules 2026, in force from 20 February 2026, require platforms to label AI-generated audio and video and embed metadata showing its source. Certain lawful takedown orders must be acted on within 3 hours, down from 36, and grievance responses within 7 days.
Sources
- FBI: cybercrime losses hit $21 billion in 2025 as AI fuels fraud, Bitdefender, 2026
- FBI IC3 2025 Internet Crime Report: 10 takeaways, SOCRadar, 2026
- Ten takeaways from IBM's 2026 Cost of a Data Breach Report, Baker Donelson, 2026
- IBM 2026 Cost of a Data Breach Report: key findings, eSecurity Planet, 2026
- Scammers siphon $25M from engineering firm Arup via AI deepfake CFO, CFO Dive, 2024
- Senior US officials continue to be impersonated in malicious messaging campaign (PSA I-051525-PSA), FBI Internet Crime Complaint Center, 2025
- Deepfake banking fraud risk on the rise, Deloitte Center for Financial Services, 2024
- IBM Report: India's Average Data Breach Cost Hits Record INR 25.5 Cr, Entrepreneur India, 2026
- Govt tightens rules on AI-generated content, mandates labels and faster takedowns, Business Today, 2026
- Verizon DBIR 2026: vulnerability exploitation, third-party risk and compliance, ComplianceHub, 2026
- Phishing Activity Trends Report, 2nd Quarter 2026, Anti-Phishing Working Group (APWG), 2026
