Intelligent Risk Control Enters the 10‑Millisecond Era

On August 18, 2022, at the IDC China Digital Finance Forum, Li Junkui, President of Ant Group's Security Business Group, delivered a speech sharing his insights on future risk‑control technology trends and the practical experience of Ant's intelligent risk‑control system.

The presentation was divided into four parts: two observations on recent risk trends, an analysis of network‑risk situations and industry technical responses, characteristics of future risk‑control technology, and Ant's risk‑control practice over recent years.

Observation 1: Since 2016, the network‑risk landscape has changed dramatically, with a surge in telecom fraud, large‑scale online gambling, and related financial losses worldwide.

Observation 2: Black‑gray market operators have upgraded their tools, moving from simple “black cards” to intelligent SMS‑blocking and AI‑driven techniques, indicating a rapid evolution of malicious AI capabilities.

These observations reveal a shift from explicit risks (e.g., account takeover) to implicit risks (e.g., scams, gambling), creating significant challenges for platforms that must balance security with user experience.

The three major challenges identified are: (1) difficulty of identification because malicious actions are performed by users themselves, rendering traditional explicit signals insufficient; (2) mind‑set confrontation, requiring strategies to change victims' perceptions rather than merely blocking transactions; and (3) an escalating offense‑defense arms race, where both sides increasingly employ AI technologies.

Looking ahead, intelligent risk control is expected to exhibit three features: privacy‑preserving end‑to‑end intelligence, machines that can understand human intent, and self‑evolution, all driven by ultra‑fast response times—ideally within 10 milliseconds.

Ant's risk‑control technology, encapsulated in the IMAGE framework (I‑M‑A‑G‑E), comprises five AI‑driven components: Interactive proactive control, Multi‑party secure control, Game‑theoretic adversarial intelligence, Graph‑based risk mapping, and Edge‑cloud collaborative control. These innovations have achieved detection accuracy improvements (e.g., 80% higher fraud identification, 40% better multi‑party risk detection, 95%+ accuracy for black‑market group detection) and maintained sub‑10 ms latency across billions of daily transactions.