The FBI’s complaint centre recorded 1,008,597 complaints for 2025 with $20.877 billion in adjusted losses. Of those, 22,364 complaints included AI-related information, representing $893.35 million.
AI-referencing complaints are therefore about 2% of the volume and roughly 4% of the losses. That is a smaller share than most commentary this year would suggest, and it is worth saying so.
The Denominator Is The Story
A million complaints and twenty billion dollars is the actual scale of the problem, and the overwhelming majority of it involves no novel technology at all — investment fraud, business email compromise, tech support scams and extortion, conducted with tools that have existed for years.
A reader following coverage alone would conclude the reverse. The reporting incentive runs toward the novel category, and the effect is that defensive attention and budget follow a small fraction of the loss.
Why The AI Figure Is Nonetheless Understated
The figure counts complaints that *included AI-related information* — which requires the victim to have known. Somebody defrauded by a cloned voice they believed was their finance director has no way to report the involvement of synthesis, because they did not detect it.
Self-reported categorisation systematically undercounts the techniques that work best, since a technique that succeeds is one the victim did not recognise. That cuts against the reassurance in the low percentage.
How We Use This Data
It is the most reliable public baseline available, and it is a complaint dataset rather than an incidence measure: it reflects what victims reported to one agency in one country. Losses recovered later are not netted off in the headline figure, and organisations frequently do not report at all.
Compiled from published complaint-centre figures and reporting, listed below. These are self-reported complaint data, not an independent measure of incidence. Corrections: corrections@forensicpost.com.