Deepfake fraud: what identifAI’s new report counts

identifAI says video and voice cloning account for most of the deepfake fraud it tracks. Its definition of an “incident” shapes what the figures can show.

identifAI, the deepfake-detection company founded by Marco Ramilli and Marco Castaldo, has published a report on deepfake fraud covering January 2020 to August 2026. According to the company:

• Video synthesis is the largest fraud modality at 42.7% of fraud incidents, followed by voice cloning at 21.9%. Together they make up 64.6%.
• The US accounts for 34.1% of incidents, ahead of India (7.7%) and the UK (6.6%).
• X (42.1%) and LinkedIn (14.5%) are the platforms where incidents most often surface.
• Identity fraud makes up 17.6% of fraud incidents. The report credits the figure that synthetic media is behind about 1 in 100 failed identity checks to Identity Week, not to its own data.

How the report counts

The report defines an incident as a deepfake used in the real world and either reported by English-language media or mentioned in a social media post. Incidents were classified by a large language model, with human review. The fraud set contains 7,926 events out of 59,747 observed, and 55.8% of the victims are classed as “other or unspecified”.

The report also describes the $40bn figure used in the press release’s headline as a projection of US fraud losses by 2027, attributed to unnamed “financial think tanks”, not a measured total.

In the company’s statement, Ramilli says deepfake detection “can no longer exist as a post-incident forensic tool” and should be built into financial workflows as an active security layer.

This article was drafted with AI assistance from identifAI’s press release and report.