Binance’s AI-driven risk systems helped protect more than 8 million users and helped prevent approximately US$4.6 billion in potential losses during the first half of 2026, as the company continues to deploy artificial intelligence across its security, anti-fraud and compliance infrastructure.
The systems cover abnormal trading activity, account takeover attempts, scams and transaction fraud, with AI increasingly embedded across the user journey wherever real-time risk decisions are required.
In H1 2026, Binance intercepted millions of scam and phishing attempts, blacklisted more than 42,000 malicious addresses, and issued more than 14,000 real-time warnings every day. Binance currently operates more than 100 AI models across its anti-fraud and anti-scam controls. The infrastructure is built, trained and supervised in-house, with human risk analysts setting thresholds, reviewing edge cases and retraining models as new scam patterns emerge.
Across fraud controls, AI models now make 80% to 90% of real-time risk decisions, while also assisting in approximately 45% of human review workflows. Human reviewers remain central to cases requiring more detailed validation and contextual judgment.
AI embedded across the user journey:
AI is deployed across identity verification, account security, payments and broader transaction protection and screening. Each action is evaluated as it occurs, with the large majority of decisions resolved automatically. This enables risk protection to operate at platform scale without slowing down legitimate users, with most users not seeing the checks taking place in the background.
Within Know Your Customer (KYC) identity verification processes, Binance’s AI-enabled review pipelines have delivered up to 100x operational efficiency gains over manual processes in specific workflows, while specialists remain involved in higher-risk cases.
Binance uses a hybrid approach combining proprietary technology with external AI and foundation models. Its in-house models are designed around risks observed specifically on the platform, while external models are used for broader reasoning tasks. The company also operates an internal Red Team that tests its defenses by examining how emerging technologies could potentially be used against the platform.
Jimmy Su, Chief Security Officer at Binance, said: “These exercises help us identify weaknesses before attackers do, validate that our controls work under realistic conditions, and continuously strengthen the people, processes and technology protecting our users. In security, you cannot simply assume your defenses will work – you have to challenge them.”
Layered protection against social engineering
Social engineering remains an area where multiple AI technologies work together, particularly in peer-to-peer trading. Binance’s in-house computer vision models detect fake
proof-of-payment images by analyzing transaction details and subtle image manipulations. AI is used for high-volume discovery and screening, while human reviewers provide more detailed validation and contextual judgment. Findings from emerging attack patterns are subsequently used to fine-tune the models, creating a continuous feedback loop in which AI provides scale and human specialists provide accuracy.
Binance also applies structured model governance throughout the AI model lifecycle, covering development, validation, deployment and ongoing monitoring.
AI supporting compliance and internal operations
Beyond direct user protection, Binance’s compliance teams use AI-assisted automation to support areas including KYC fraud detection and transaction-monitoring execution.
More than 24 AI initiatives have been deployed across user onboarding, screening escalations and partner due diligence.
AI is also being adopted across Binance’s internal operations. The company’s internal agentic tool currently has approximately 72% uptake across teams, supported by company-wide training, prompt-engineering programs and structured oversight.
Binance says its AI systems operate within a privacy-first framework built around data minimization, purpose limitation and user-rights safeguards. These principles are incorporated into how AI models are developed and deployed, with the objective of protecting users from financial abuse and harm without compromising data privacy.
As increasingly sophisticated forms of deception become easier and cheaper to produce, Binance continues to invest in a combination of AI-driven scale and human judgment, including retraining models, refining thresholds and keeping experienced analysts involved in cases requiring more detailed review.
