On April 6, the FBI released its 2025 Internet Crime Report, marking 25 years of the Internet Crime Complaint Center. The headline numbers are staggering: nearly $21 billion in losses, over one million complaints, and a 26% increase in losses over 2024. But the development that matters most sits in the middle of the report.

For the first time in its history, IC3 dedicated an entire section to artificial intelligence in cybercrime. That section documents 22,364 AI-related complaints with losses of $893 million. The FBI now treats AI as a category of criminal tooling in its own right.

This matters for anyone working in scam prevention and defense. The report confirms what structured classification frameworks have been built to capture: a scam is a layered operation, a sequence of tactics that span access, manipulation, deception, and extraction. The IC3 data, when mapped against a lifecycle-based taxonomy like the SC&MS framework, reveals patterns that aggregate statistics alone cannot show.

The Numbers

IC3 received 1,008,597 complaints in 2025, up from 859,532 in 2024. Total reported losses reached $20.877 billion. Cyber-enabled fraud accounted for 452,868 of those complaints and $17.7 billion in losses, representing 85% of all financial losses reported to IC3.

Investment fraud remains the single largest driver, accounting for $8.65 billion in losses. Business email compromise followed at $3.05 billion, then tech support fraud at $2.13 billion.

AI as Fraud Infrastructure

The 22,364 AI-related complaints break down across multiple scam types. The FBI identified five primary ways AI is being deployed in fraud operations:

  • Investment scams ($632 million): AI-generated celebrity and CEO endorsement videos, mass-generated individualized conversations with prospective victims, and fabricated professional-looking investment groups on social media.
  • Business email compromise ($30.2 million): Voice cloning used to authorize wire transfers, and AI chat generators creating convincing executive-impersonation emails.
  • Romance and confidence scams ($19 million): AI-generated chat scripts to maintain believable conversations at scale, and fake social profiles created with synthetic media.
  • Grandparent and distress scams ($5 million+): Voice cloning technology used to mimic the sound of a family member in distress, evolving beyond grandparents to other relationships and emergency scenarios.
  • Employment scams (almost $13 million): Voice spoofing and video deepfakes during remote job interviews, where lip movements do not match audio. The primary goal appears to be gaining access to corporate networks rather than direct financial theft.

Mapping IC3 Findings to the ABCDEF Lifecycle

These AI-enabled tactics do not exist in isolation. Each one maps to specific stages of the scam lifecycle. When viewed through the SC&MS framework, the FBI's findings show how AI amplifies existing techniques. One stage, Coercion, is missing from the headings below by design; the Elder Fraud section explains why.

Access (A-stage)

The IC3 report notes that crypto investment scammers "typically initiate contact through text messages, social media sites, advertisements, or dating applications." These map directly to SC&MS access vectors: A002 (text message), A009 (social-media direct message), A013 (online advertisement), and A007 (dating-app message). AI does not change how scammers reach people. It changes what happens after contact is made.

Bait (B-stage)

Investment fraud ($8.65 billion) and romance scams ($929 million) dominate the loss figures. The lures map to B010 (investment opportunity), B015 (romantic interest), and B017 (job opportunity). The lures grow more personalized and harder to detect; the bait categories underneath them do not move.

Deception (D-stage)

This is where AI has the most direct impact. The FBI's findings map to several SC&MS deception techniques:

  • D012.001 (voice cloning): Used in BEC wire fraud, grandparent scams, and distress calls
  • D012.002 (video deepfake): Used in employment interview fraud and celebrity endorsement videos
  • D013 (fabricated persona): AI-generated social profiles at scale for romance and investment scams
  • D014.001 (fabricated endorsements): AI-generated celebrity and CEO endorsement videos for investment clubs

Exploiting Trust (E-stage)

The IC3 report describes scammers being "introduced to investment groups representing themselves to be knowledgeable industry insiders." This maps to E005 (public figure impersonation), E012 (manufactured relationship), and E013 (trusted intermediary referral). AI-generated content makes these impersonations more convincing, but the trust mechanisms themselves predate AI.

Financial Gain (F-stage)

Cryptocurrency dominates extraction. Across 181,565 crypto-related complaints, losses reached $11.37 billion. The report shows cryptocurrency as the top reported transaction type in investment fraud (72%), tech support fraud (43%), and government impersonation scams (40%). This maps to F005 (cryptocurrency payment) and F002 (fraudulent investment deposit). Crypto ATM losses jumped 58% year over year to $389 million, with victims over 60 accounting for $257 million of that total.

Elder Fraud

Americans over 60 filed 201,266 complaints and reported $7.748 billion in losses, a 59% increase over 2024. The average loss was $38,500, and 12,444 complainants lost more than $100,000 each. Investment fraud and tech support fraud were the leading categories for this age group, with crypto investment scams accounting for $2.76 billion in losses among victims over 60 alone.

These numbers point to a concentration of coercion and trust-exploitation techniques targeting older adults. The IC3 data does not break down which specific pressure tactics are used by age group, but the loss severity suggests that long-duration, high-manipulation operations like romance-baited investment scams are disproportionately affecting this population. Coercion is the one lifecycle stage this analysis leaves unmapped for that reason: the report's aggregate data names the pressure without identifying the specific techniques, so no C-stage mappings are claimed here.

Recovery Scams

Recovery scams generated 10,516 complaints and $1.4 billion in losses. These are scams where criminals contact previous victims, claiming to help recover their stolen funds. The FBI notes an increase in scammers impersonating government officials and recovery firms. In the SC&MS framework, this is a full second pass through the lifecycle: a new access vector (A003, phone call, or A001, email), a new bait (B021, lost or stolen fund recovery), and renewed trust exploitation through authority impersonation (E002).

Reporting Itself Becomes the Lure

Late last month, before this analysis was published, the FBI issued a public service announcement (July 20, 2026) warning of a scheme that activates at the moment a victim tells their scammer they intend to report the fraud. Criminals had built a deepfake video impersonating a senior FBI official, paired with a spoofed version of the ic3.gov reporting site, so the scammer can step in as the reporting channel itself. The victim believes they are finally escaping the scam. They are being handed back to it.

In the SC&MS framework, this is trust exploitation folded back on the escape route: authority impersonation (E002) built not on a generic government seal but on a synthetic likeness of a named official and a cloned reporting portal, layered on top of the original operation. The report documents the same voice-cloning and video-deepfake techniques as investment and employment fraud tools (D012.001, D012.002). Here they are aimed at the reporting system a victim reaches for when they realize they have been had.

It is a fitting illustration of the report's own lesson. The technique inventory is stable. What changes is how convincingly, and against whom, each technique gets deployed.

What This Means for Scam Classification

The IC3 report reinforces a core premise of lifecycle-based classification: scam operations are not defined by a single technique. They are composed of techniques drawn from multiple stages, combined and adapted for different targets and contexts. AI is not a new scam type. Instead, it is an accelerant that makes existing deception techniques cheaper and harder to detect.

A classification system that treats "AI scam" as a category will struggle to capture what is actually happening. A system that tracks voice cloning as a deception technique, romance cultivation as a trust mechanism, and cryptocurrency as an extraction method can map the same scam operation at each stage and identify where defenses might intervene.

The full 2025 IC3 report is available from the FBI. The SC&MS taxonomy is available in the Explorer and on GitHub.