Americans lost nearly $893 million to scams powered by artificial intelligence in 2025, according to the FBI’s Internet Crime Report, which for the first time broke out AI-enabled fraud as its own category. The bureau logged 22,364 complaints tied to AI tactics within a broader year of cyber-enabled crime that totaled roughly $21 billion in losses. Separately, the Federal Trade Commission reported that imposter scams alone cost people $3.5 billion in 2025, a figure that overlaps with and amplifies the FBI’s AI-specific count.
Why the $893 million AI fraud total signals a structural shift
The decision to carve out an AI section inside the annual Internet Crime Report was itself a signal. Before this year, AI-assisted fraud was folded into broader categories such as business email compromise, romance scams, and investment fraud. By isolating 22,364 complaints and nearly $893 million in losses, the FBI acknowledged that generative tools have become a distinct force multiplier for criminals rather than a marginal add-on to existing schemes. In a separate press statement on cryptocurrency and AI scams, the bureau framed these tools as accelerants that help criminals scale operations and reach victims they might otherwise miss.
The average loss per AI-related complaint works out to roughly $39,900 when dividing the total by the complaint count. That figure sits well above the per-complaint average across all cyber-enabled crime categories when measured against the $21 billion overall total and the hundreds of thousands of complaints the bureau typically receives each year. The gap suggests that AI-enhanced scams do not simply produce more attempts; they produce more successful and more expensive ones. Testing that hypothesis precisely would require clearance-rate and recovery data the FBI has not yet published, but the raw ratio already points in that direction.
The FTC’s parallel data reinforces the pattern. The commission found that people reported losing $3.5 billion to imposter scams in 2025, with phone calls, text messages, and social media all serving as common entry points. In its release on reported imposter losses, the agency emphasized that scammers increasingly pose as government officials, business representatives, or family members in crisis. Impersonation fraud is the category most directly supercharged by generative AI, which can clone voices, generate realistic video, and produce grammatically flawless messages in any language. The FTC figure is not limited to AI-driven cases, but the overlap between the two agencies’ data sets shows that impersonation at scale is now the dominant fraud vector in the United States.
How generative tools changed the mechanics of fraud
The FBI’s Internet Crime Complaint Center had already documented the specific techniques before the annual numbers confirmed their scale. A public service announcement published in December 2024 warned that criminals were using generative AI to craft convincing phishing messages, improve grammar and syntax across multiple languages, and build fake personas for social engineering campaigns. That IC3 advisory on generative AI in financial fraud described a toolkit that lowers the skill floor for would-be scammers while raising the ceiling of believability for victims.
A second IC3 advisory, issued in 2025, described a more alarming application: criminals using altered proof-of-life media in virtual kidnapping schemes. In these cases, scammers generate or manipulate audio and video to convince a victim that a family member has been abducted, then demand ransom payments. The tactic exploits the same deepfake technology that powers entertainment filters and corporate video tools, repurposed for extortion.
These advisories sketch a clear progression. Generative text made email and messaging scams harder to spot by eliminating the spelling errors and awkward phrasing that once served as red flags. Synthetic voice cloning made phone-based impersonation more convincing, allowing criminals to mimic a loved one or company representative with chilling accuracy. Deepfake video added a layer of false proof that can override a victim’s skepticism in real time, especially when delivered under pressure with urgent demands for payment.
Each step widens the gap between what automated defenses can catch and what a targeted individual can detect on their own. Spam filters and fraud-detection algorithms trained on older patterns struggle to keep up with content that looks and sounds indistinguishable from legitimate communication. At the same time, victims are asked to make faster decisions, often on mobile devices, where contextual cues are easier to miss.
Gaps in the data and what to watch next
The FBI’s new AI category is a useful starting point, but it leaves significant questions unanswered. The bureau did not publish a breakdown showing how much of the $893 million came from voice cloning versus deepfake video versus AI-generated text. Without that granularity, banks, telecom companies, and platform operators cannot easily prioritize which detection tools to deploy first or which channels-phone, email, messaging apps-pose the greatest immediate risk.
Victim demographics are also absent. Neither the FBI nor the FTC releases from 2025 specify age groups, income brackets, or geographic concentrations for AI-driven fraud. Older adults have historically been overrepresented in imposter scam losses, but generative AI may be shifting the target pool by making scams more plausible to younger, digitally fluent victims who are accustomed to seamless video calls and automated customer service. That question cannot be answered with the data currently available, leaving policymakers and consumer advocates to extrapolate from partial trends.
Complaint resolution rates and recovery amounts remain unpublished as well. The $893 million figure represents reported losses, not net losses after law enforcement intervention or bank clawbacks. If recovery rates for AI-enhanced fraud are lower than for traditional scams-because transactions move through cryptocurrency, overseas payment processors, or rapidly cycled accounts-the effective damage is even larger than the headline number suggests. Conversely, if banks and platforms are quietly improving their ability to reverse fraudulent transfers, the raw totals may slightly overstate the long-term financial hit while understating the emotional toll.
These gaps matter because they shape how resources are allocated. Law enforcement agencies must decide whether to invest more heavily in digital forensics, victim outreach, or international partnerships. Financial institutions are weighing whether to harden identity verification at the cost of customer friction or to focus on back-end anomaly detection. Without better visibility into which AI techniques cause the most harm, those decisions risk missing the mark.
What consumers and institutions can do now
For anyone who suspects they have been targeted, the federal intake points remain active. The FTC accepts fraud reports at reportfraud.ftc.gov, and the FBI’s Internet Crime Complaint Center continues to collect detailed complaints that feed into its annual reporting. Submitting a report does not guarantee recovery, but it helps investigators link cases, identify emerging AI tactics, and pressure intermediaries that enable the scams.
Consumers can also adjust their own verification habits in light of AI’s capabilities. Treat unexpected requests for money or sensitive information-especially those delivered via voice or video-as unverified, even if they appear to come from a trusted person or institution. Whenever possible, use a second channel to confirm the request, such as calling a known phone number or logging directly into an official website rather than clicking a link. Families may benefit from prearranged “safe words” or questions that are harder for an impostor to guess from social media alone.
Institutions, meanwhile, are being pushed to rethink authentication. Banks and service providers are experimenting with layered checks that combine device fingerprints, behavioral patterns, and out-of-band confirmations before processing high-risk transactions. Employers are revisiting internal approval workflows for payments and data access, recognizing that a convincing AI-generated email or voicemail can defeat single-person controls.
For those who want to track how these threats evolve, the FBI encourages the public to sign up for email alerts that include future Internet Crime Reports and public service announcements. As generative AI tools continue to advance, the numbers from 2025 are unlikely to be a peak. They are better understood as a baseline for a new era of fraud in which the line between authentic and synthetic communication grows thinner-and the cost of confusion keeps rising.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
