Sumsub has found that more than half of businesses have suffered fraud involving AI agents, while just 12% are able to suspend or revoke an agent’s access, according to its sixth annual Identity Fraud Report 2026–2027.
The research shows the global identity fraud rate dropped 28% year on year to 1.6%. That fall masks a tougher reality, however. Deepfakes climbed 31% worldwide, and attacks are becoming more organised, more AI-driven and harder to spot.
Sumsub argues that fewer fraudulent attempts should not be read as lower exposure, since criminals are using sophisticated tools, connected networks and AI-powered workflows to target businesses.
Agentic fraud stands out as a major new threat. Sumsub’s Fraud Exposure Survey 2026 found that most firms have already been affected by fraud linked to AI agents. Only 37% of companies monitor who deploys these agents.
Organisations are increasingly letting software act for people without any way to confirm its identity, check its authority or remove its permissions.
Coordinated fraud networks are another growing concern. The survey found that 74% of companies fail to detect network-level fraud signals, and just 21% use fraud network analysis. A third of businesses that experienced fraud said at least 25% of their confirmed cases were tied to a broader network, linked through shared devices, contact details, payment instruments, locations, IP patterns or document templates.
Sumsub’s own data suggests roughly 2% of verified users later become associated with a fraudulent cluster. Some of these clusters span several countries and involve thousands of users.
Source link







