Artificial intelligence has lowered the cost of making a lie look and sound believable. Criminals can now produce a familiar voice, a polished investment pitch or a convincing fake identity in minutes. New FBI data show that complaints specifically involving AI generated losses on a scale approaching $900 million in a single reporting year.
The $893 million figure comes from victim reports
The FBI’s Internet Crime Complaint Center, known as IC3, collects reports from individuals and organizations. Its annual totals reflect money victims say they lost, not every attempted scam and not a complete measure of unreported crime.
In the FBI’s April 2026 release, the bureau said 22,364 complaints involving artificial intelligence accounted for nearly $893 million in reported losses during 2025. The AI section appeared for the first time in the report’s nearly 25-year history.
The larger cybercrime picture was even more severe. IC3 received more than one million total complaints and reported nearly $21 billion in overall losses. Americans over 60 reported about $7.7 billion, making older adults a particularly costly target pool.
AI improves the disguise rather than inventing the scam
Many underlying schemes remain familiar: investment fraud, business-email compromise, tech-support fraud and family-emergency calls. AI adds speed and personalization. A fake profile can carry realistic photos and text, while cloned audio can imitate a relative or executive.
The 2025 IC3 report describes synthetic media, false documents and impersonation as tools used to build trust or urgency. The technology makes separate pieces of a story agree with one another, which can defeat a quick visual check.
Voice cloning turns public audio into a weapon
Scammers do not need a long private recording to begin imitating a voice. Public videos, voicemail greetings and social-media clips can provide useful samples. The resulting call may claim a child or grandchild has been arrested, injured or kidnapped.
The FTC has warned that AI can strengthen family-emergency schemes. The safest response is to end the unexpected call and contact the person through a known number. A family code word can add another check, but it should not be shared publicly.
Payment instructions expose the criminal script
Pressure to act immediately is a core warning sign. So are demands for cryptocurrency, gift cards, cash couriers, wire transfers or a transfer to a supposedly safe account. These methods make recovery difficult and separate the victim from ordinary fraud controls.
A real company or agency can tolerate independent verification. Contact information supplied by the caller should not be used for that check. Statements, official websites and phone numbers already known to be genuine provide a safer route.
A pause remains the strongest defense
The FBI urges potential victims to “Take a Beat” before sending money or information. That pause creates time to inspect the story, involve another person and confirm identities outside the channel the scammer controls.
Victims should contact the bank or payment service quickly, preserve messages and transaction details, and report the incident to IC3 and the FTC. Fast reporting cannot guarantee recovery, but it can help stop transfers and connect related cases.
Businesses face a parallel threat when criminals imitate an executive and request a wire, payroll change or confidential file. A second-channel approval rule can require staff to confirm unusual transactions through a known phone number or internal system. The control works even when the email, voice and video all appear authentic.
Public figures and companies can make impersonation harder to exploit by publishing official contact routes and warning customers that support teams will never request certain payment methods. Platforms can label synthetic content, but detection tools will remain imperfect as generation improves. Verification procedures are more durable than trying to recognize every fake by visual flaws.
The loss figure may rise partly because reporting categories are improving. The first dedicated AI section gives investigators a baseline, but year-to-year comparison will require consistent definitions. A complaint involving cryptocurrency, investment fraud and a cloned voice may belong to several categories at once.
The FBI’s total does not show that AI alone stole the money. It shows that AI has become part of a costly fraud ecosystem already built around pressure and impersonation. The technology can fake a voice; it cannot prevent an independent callback, a second opinion or a refusal to move money on command.
The classification problem also limits claims about which tool caused a loss. A synthetic voice may open the conversation, but the decisive steps can be a fake investment platform, social engineering and an irreversible transfer. Prevention therefore cannot stop at deepfake detection; it must protect the payment and account-recovery stages that criminals ultimately need.
Organizations can rehearse those controls before an emergency. Written approval thresholds, known callback numbers and a ban on changing payment instructions solely by email make an urgent request easier to challenge. The process should apply to senior leaders as well as staff because a convincing imitation often borrows authority from the top of an organization.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
More from Morning Overview
- Two Ford recalls now carry a rare “do not drive” warning, so owners should check their VIN.
- A Colorado wildfire has exploded past 87,000 acres with no containment
- A wildfire near Ouray swelled past 18,000 acres, putting Colorado towns on alert
- The Pentagon’s newest UFO files describe a fish-scaled, potato-shaped object on camera
