AI in Cybercrime: What the FBI's 2025 Complaint Data Show

The FBI's Internet Crime Complaint Center (IC3) counted 22,364 complaints in 2025 that referred to artificial intelligence, with adjusted losses of USD 893,346,472. That is about 2 percent of the 1,008,597 complaints the center received that year and about 4 percent of the USD 20.877 billion in total reported losses. The report calls its own picture incomplete, because victims often cannot tell whether a scammer used AI. It is the latest edition available as of October 2026; it appeared in April 2026.

What counts as an AI-related complaint

IC3 tags each complaint with a crime type, and it can also add a descriptor that records the medium or tool involved. The "AI Related" descriptor is applied when the information reported "contains a reference to artificial intelligence." The tag therefore reflects what a victim or investigator wrote down, not a forensic finding about how a message was made. A victim who received a polished email may never learn that a chatbot wrote it, and one who suspects AI without evidence may still mention it. Both effects distort the count in opposite directions, and the report itself says that many victims do not realize how much AI may have been involved.

Where the AI mentions cluster

The report lists AI references by crime type. Investment scams lead by a wide margin, followed by extortion and personal data breaches.

Complaint counts come from the report's chart; loss figures come from its narrative text, which covers only some categories.

How the FBI says the tools are used

The report describes specific patterns. In business email compromise, chat generators write official-sounding messages that imitate a chief executive, and voice cloning can be used to ask for a wire payment. In romance scams, AI chat generators supply scripts and fake profiles, and the same technology can mimic a relative's voice in "distress" scams, which drew reported losses of more than USD 5 million in 2025. In employment fraud, the lip movements of a job applicant on camera may not match the audio, and the report notes that the goal there often appears to be access to private computer networks rather than money. In investment scams, subjects use AI to generate thousands of conversations that each look different, and "investment clubs" use generated videos and voices of celebrities or executives to lend credibility.

Reading the numbers without overstating them

Two calculations help put the AI figures in scale. Dividing AI-related losses by AI-related complaints gives roughly USD 39,900 per complaint, against roughly USD 20,700 per complaint across all of IC3 (both calculated from the report's totals). The AI-tagged group therefore skews toward larger losses, which fits the heavy share of investment cases. In investment fraud, AI-tagged losses of more than USD 632 million sit inside total investment losses that the report puts above USD 8 billion, so under 8 percent of that category carries an AI reference. The report reads that gap as evidence of under-recognition, which is a plausible explanation but not something the data can prove.

What comes next in the data

The 2025 report is the first IC3 annual report to give AI its own section, and no later edition exists yet. A report on 2026 complaints has not been published as of October 2026. The FBI released the previous annual report on April 23, 2025, and the 2025 edition followed in April 2026, so a spring 2027 release would fit the pattern, though no date has been announced. Until then, year-over-year comparisons of AI-tagged losses are not possible from IC3 data.

Practical checks that follow from these patterns

• Confirm any request to move money through a second channel that the requester did not supply, such as a known phone number.
• Treat a video call with a celebrity, executive, or relative as a claim to verify, not as proof of identity.
• Agree on a family or workplace code phrase for emergencies, so a cloned voice cannot pass on sound alone.
• Slow down when a message creates urgency about a transfer, a trade, or a login.
• Report incidents to IC3, because the center's counts depend on victims filing complaints.

These steps are general fraud-prevention practice drawn from the patterns above and are not quoted from the report.

FAQ

Does the report show that AI scams are growing?

The AI section of the 2025 report, still the latest edition as of October 2026, gives 2025 figures only, with no year-over-year AI comparison. Total losses did rise from USD 16.6 billion in 2024 to USD 20.877 billion in 2025, a 26 percent increase, but the report does not attribute that rise to AI.

Are the USD 893 million losses all caused by AI?

No. They are losses in complaints that mention AI. Many of those scams, such as investment fraud, existed long before generative AI and could have run without it.

Do these numbers cover Canada?

No. IC3 is a U.S. federal reporting channel, and its figures describe complaints filed there. Fraud reported to Canadian authorities is counted separately under different rules, so the two sets cannot simply be added together.

Can people tell when a voice or video is AI-generated?

The FBI says AI-enabled synthetic content is becoming harder to detect and easier to make. Mismatched lip movement or audio that does not line up with actions are signs the report mentions in employment cases, but they are not reliable on their own.

Related reading

• Whether attackers or defenders gain more from AI
• Guidance from cyber agencies on securing AI systems
• Two studies on security flaws in AI-written code
• Smart contract audits and exploits

More in AI Policy and Ethics.

Last verified: October 2026

This page reports statistics from a law enforcement document. It is general information about fraud and not legal or financial advice.

Sources

• 2025 Internet Crime Report, Federal Bureau of Investigation, Internet Crime Complaint Center, 2026
• FBI Releases Annual Internet Crime Report, Federal Bureau of Investigation, 2025