Quick answer: The FBI Internet Crime Complaint Center's 2025 Annual Report added AI-related as a crime descriptor for the first time in the agency's 25-year history. The category captured 22,364 complaints and $893.3 million in losses in its first year. AI-enabled fraud grew 1,210% in 2025 versus 195% for non-AI fraud. Seniors accounted for $352 million of the losses.
For 25 years the FBI's Internet Crime Complaint Center tracked online fraud in categories like wire fraud, BEC, identity theft, and investment scams. In 2025, for the first time, the IC3 Annual Report added a new descriptor: "AI-related." The category exists now because the underlying conduct can no longer be understood without it.
This post walks through the headline numbers, the per-category breakdown, and what each tells you about which AI fraud vectors are currently doing the most damage.
For the broader technical grounding on how synthetic media is generated, see the pillar guide on what a deepfake actually is.
1,210%
The year-over-year growth rate for AI-enabled fraud reported to the IC3 in 2025, compared with 195% growth for non-AI fraud. AI-powered schemes are now scaling approximately six times faster than every other kind of online fraud the FBI tracks.
Source: FBI Internet Crime Complaint Center 2025 Annual Report.
What the Report Actually Adds
The IC3 has tracked online fraud since 2000. For the first 24 of those 25 years, "AI" was not a crime descriptor on the complaint form. The 2025 report introduces it because the agency can no longer cleanly categorize what victims are reporting without it. A romance scam that uses a cloned voice is, simultaneously, a romance scam and an AI-enabled romance scam, and the AI component changes both the victim count and the loss profile. The new descriptor is how the data infrastructure catches up.
The 22,364 complaints filed under the AI-related descriptor in 2025 are a floor, not a ceiling. The IC3's own methodology note acknowledges that AI tagging is voluntary, that many complaints involve AI components the complainant did not identify as such, and that the actual AI-related share of fraud is likely meaningfully larger than the descriptor-tagged subset. The number captures self-identified AI fraud and undercounts the broader category.
The dollar figure, $893.3 million, is similarly conservative for the same reasons. The agency provided the growth comparison (AI-enabled +1,210% vs non-AI +195%) precisely because the trajectory matters more than the level. A category that adds an order of magnitude of loss volume in one year is the category to watch.
The Per-Category Breakdown
The AI-tagged complaints clustered into five recognizable patterns.
AI Investment Scams: $632 Million
The single largest dollar category in 2025. AI is used to generate videos and voices of well-known figures, executives, and athletes endorsing fraudulent investment platforms. The mechanism is the one documented in detail in our coverage of celebrity crypto-endorsement deepfakes and the Bombay Stock Exchange CEO deepfake stock-pump scam. Victims see what looks like a credible endorsement, follow the link to a polished-looking platform, deposit funds, and find the withdrawals blocked or the platform gone.
The $632 million figure is what the FBI captured under the AI-investment subset specifically. Adjacent categories (the Truman Show fake-investment-group pattern, the fake AI investment group playbook) push the real number higher.
Voice Cloning and Family-Distress Calls: Embedded Across Categories
The IC3 report calls out voice cloning specifically: cloned voices of family members in distress or emergency scenarios, used to extract cash transfers and gift cards. The pattern is documented in our voice cloning detection guide and the family-protection playbook.
The cloning threshold has collapsed: McAfee documented in 2023 that three seconds of source audio is enough to produce a working clone. The IC3 data shows that operators are now using this at scale. The behavioral defense (callback to a known number) remains the only thing that does not erode as model quality improves.
Voice cloning does not appear in the IC3 report as a separate dollar line. It cuts across several other categories: it shows up in romance scams when a "boyfriend" or "girlfriend" places an audio call, in elder fraud when a "grandchild" claims to need bail money, in business email compromise when an executive's voice authorizes a wire transfer, and in government-impersonation calls when a "federal agent" demands payment. Each of those categories has its own dollar line in the IC3 report; the voice-clone component is the technology being used across them, not a standalone line item. That is why the IC3 documents voice cloning narratively in the report's methodology discussion rather than aggregating it as a separate sum.
Senior-Targeted Fraud: $352 Million of the $893M
Forty percent of AI-related losses came from victims aged 60 and over. The category cuts across all the others (seniors are over-represented in voice-clone-family-emergency, investment-scam, and government-impersonation cases). The targeting is deliberate: older adults are more likely to take an unexpected call, more likely to accept urgency framing, and more likely to have liquid retirement savings within reach.
For families coaching an older relative through the threat surface, the voice cloning family-protection guide and the Facebook detection guide's parent-help section are the practical references.
Government and Official Impersonation: Doubled in 2025
Per related Nextgov reporting on the same FBI release, government-official-impersonation complaints doubled in 2025. The scam: a call or message that appears to come from the IRS, the Social Security Administration, a local police department, or a federal agent, often with a cloned voice and a spoofed caller ID, demanding immediate payment to avoid arrest or to settle a fabricated case.
The defense pattern is the same as for the family-distress version: end the call, look up the official number for the agency, call back through that number. No real federal agency conducts business by demanding immediate payment over the phone.
CSAM Referrals to NCMEC: 5,700+ in 2025
The IC3 referred more than 5,700 submissions involving minors to the National Center for Missing & Exploited Children in 2025. The CSAM volume separately documented by NCMEC's own CyberTipline is in the tens of millions of files per year. AI-generated CSAM is a recognized and growing subset, and the federal response has been visible: the first conviction under the TAKE IT DOWN Act in April 2026 (covered in our piece on the early enforcement record) involved more than 700 AI-generated images of minors.
For parents whose child has been targeted, the parents' guide to deepfake nudes in schools covers the school-specific reporting flow and the NCMEC pathway.
Methodology: What the Numbers Cannot Capture
The 22,364 AI-tagged complaints and the $893.3 million loss figure are the floor, not the ceiling, and the IC3 says so directly in the report. Three structural reasons the actual numbers are larger.
First, the AI descriptor is voluntary. The IC3 complaint form added a checkbox for AI involvement; complainants who do not identify the AI component (because they did not know the call used a cloned voice, did not know the video was a deepfake, did not realize the "investor" they were talking to was a synthetic persona) get categorized under the underlying scam type without the AI tag. The voice-cloned grandparent-scam call gets logged as elder fraud, not as AI-related elder fraud, unless the victim or the family identifies the cloning later. Most do not.
Second, the report is the IC3 sample, not the universe. The FBI's IC3 captures the fraction of US online fraud that victims report. AARP's 2025 survey work suggests roughly 15 percent of voice-cloning victims report to any authority, citing shame and disbelief that anything can be done. Whatever the IC3 sees is a multiple-times-undercounted slice of what is happening. The 1,210% growth rate is the trajectory of the visible portion; the unseen portion is presumably tracking the same direction.
Third, the categories are not mutually exclusive. A romance-scam victim who was groomed by AI-generated images, kept on the hook by AI-cloned voice calls, and finally extracted from by a fake AI-generated investment platform shows up in the IC3 dataset once, under whichever primary scam type the complaint form forced them to pick. The AI components stack inside a single victim journey; the dollar figure does not double-count them.
What the report does well is establish a primary-source federal floor. What it cannot do is tell you the ceiling. The honest read is "the real numbers are larger, and the IC3 has now publicly committed to tracking the trajectory."
What the Numbers Mean for Consumers
The report is a primary-source confirmation of patterns consumers have been seeing. Three things follow.
The AI fraud category is no longer a future-tense concern. When a federal agency adds a descriptor after 25 years of not needing one, the agency is signaling that the phenomenon is now distinct, large, and persistent enough to warrant its own tracking. Treating AI-enabled fraud as an emerging risk is no longer accurate. It is a current threat with a current dollar floor.
The fastest-growing categories deserve the most attention. AI-tagged investment scams ($632M), senior-targeted scams ($352M), and government impersonation (doubled in one year) are where the dollar weight and the trajectory are clearest. If you have a relative in any of those exposure windows, the detection guides above are the relevant references.
The undercounting matters. The 22,364 complaint count and the $893M dollar figure are the floor that the FBI's own methodology acknowledges. The IC3 captures what victims report and identify as AI. The broader category, including unidentified-as-AI fraud, voluntary-non-report cases, and the cases that fall under non-AI descriptors despite having AI components, is meaningfully larger. The trajectory the report shows is the conservative read of a faster-growing reality.
If you want to check whether a specific video or social-media account that targeted you has been flagged by the Ledger community, paste the URL below.
Think you found an AI video?
Paste the URL and let the Ledger community verify it. Free.
The Bigger Pattern
The FBI's choice to add "AI-related" as a crime descriptor is the official record catching up to what the consumer-protection ecosystem has been documenting independently for two years. AARP, McAfee, NCMEC, Group-IB, Check Point, and the platform transparency reports have all been pointing in the same direction. The IC3 numbers now anchor the conversation in primary-source federal data.
The defense framing does not change with the data; it just gets confirmed by it. The detection skills (visual, audio, behavioral), the verification habits (callback, side-channel, regulator-registry check), and the reporting flows (FTC, FBI IC3, NCMEC, platform-specific takedown) are the same ones the smaller-scale 2024 reports recommended. The IC3 report is the validation that those defenses are needed at scale, not the introduction of new ones. The right read is "the official record now matches what the detection guides have been describing."
Related Posts
- What Is a Deepfake? A Plain-English Guide for Social Media Users: the technical foundation that explains how the underlying generation makes this fraud category economically viable
- AI Voice Cloning Scams Hit 1 in 10 Americans. Here Is How to Protect Your Family.: the family-protection sibling for the senior-targeted voice-cloning category that drove a large share of the $352M senior loss figure
- Deepfake Romance Scams Cost Americans $1.1B in 2025. Here Is How to Spot One.: the romance-scam sibling for the relationship-built-on-AI-trust category that overlaps significantly with the AI investment-scam dollar weight
- That Celebrity Crypto Video Is Probably a Deepfake. Here Is How to Tell.: the AI investment-scam detection guide for the $632M largest dollar category in the report
