INTERPOL Survey Links AI to 55% of Reported African Cybercrime

The report describes AI as a tool for automating reconnaissance, phishing, extortion and evasion. It links generative systems to convincing impersonation, deepfakes, personalized phishing messages, synthetic identities and malware designed to avoid signature-based detection.

Online scams remained the most frequently reported cybercrime category. The assessment says mobile-money fraud and ransomware targeting infrastructure were prominent in East Africa; business email compromise and romance fraud were prevalent in Central and West Africa; and Southern Africa's high connectivity attracted globally active threat groups. Seventy-two percent of surveyed countries reported scam centers within their borders, concentrated most heavily in Southern and West Africa.

TrendAI, one of the private-sector data contributors, recorded about 600,000 sextortion detections in Africa during 2025. INTERPOL also documented criminals combining real personal data with fabricated details to create synthetic identities capable of bypassing some biometric checks, opening bank accounts, obtaining mobile loans and registering SIM cards.

Four INTERPOL-coordinated operations produced 1,605 arrests and the seizure of more than 5,300 devices. The report lists recoveries of more than $4.3 million, about $2.8 million and about $3 million for three of those operations; for Serengeti 2.0, it reports networks tied to $300 million in losses rather than recovered funds. Seventeen countries enacted or amended cybercrime legislation in 2025, and Senegal launched an online reporting platform for violations affecting children.

The report still identifies fragmented laws, limited AI readiness in many law-enforcement agencies and slow information exchange among banks, telecommunications providers and investigators. Its recommendations include standardized digital forensics, stronger cross-border cooperation, AI literacy for investigators and formal public-private partnerships.

For security and ML teams, the practical signal is not simply that fraudulent messages may read better. Automated variation, synthetic media and blended identities weaken controls that rely on writing quality, a familiar voice or face, or one biometric signal. Defenses need transaction context, device and account history, step-up verification, anomaly detection, and rapid cross-organization sharing. The survey's scope and reporting limitations should remain attached to the headline number when it is used in risk models or executive reporting.