Key topics of the post:

  • In the first half of 2026, Binance’s AI-driven risk systems helped protect more than 8 million users and prevent approximately US$4.6 billion in potential losses.

  • Currently, AI underpins 80% to 90% of real-time risk decisions in identity verification, account security, payments, and transaction screening, with human reviewers handling extreme cases that require more refined judgment.

  • More than 100 AI models power anti-fraud controls to detect scams, forged documents, and large-scale social engineering attempts.

A forged payment screenshot or a tampered identity document can now be generated in seconds, and a scam script can be customized for a single target at virtually zero cost. As deception becomes cheaper to produce, the only realistic way to keep up is to continuously assess risk, at scale and in real time.

This is where AI became central to how Binance protects its users. In the first half of 2026, Binance’s AI-powered risk systems helped safeguard more than 8 million users and prevented approximately $4.6 billion in potential losses, covering abnormal trading activity, account intrusion attempts, scams, and transaction fraud.

This blog details the numbers: where AI operates across the user journey, how internal and external models work together, why human reviewers remain central to the cycle, and how the approach goes beyond loss prevention to achieve compliance and internal operations.

Binance turns more than 100 AI models into real-time user protection

In the first half of 2026, Binance intercepted millions of scam and phishing attempts, put more than 42,000 malicious addresses on the blocklist, and issued more than 14,000 real-time alerts per day.

These results come from an AI infrastructure that Binance develops, trains, and oversees internally. Currently, we run more than 100 AI models in our anti-fraud and anti-scam controls, with human risk analysts setting thresholds, reviewing edge cases, and retraining models as new scam patterns emerge.

In KYC (identity verification), our AI-enabled review pipelines delivered up to 100x gains in operational efficiency versus manual processes in specific workflows, while keeping specialists in the loop for higher-risk cases.

Binance builds AI risk protection across the user journey

AI is embedded throughout the user journey whenever a real-time risk decision needs to be made. This includes identity verification, account security (such as detecting signs of account intrusion), payments, and broader transaction protection and triage.

Each action is evaluated at the moment it happens, and the vast majority is resolved automatically. That’s what enables protection to operate at platform scale without slowing down legitimate users, since most people never see the checks running in the background during their activity.

A hybrid stack of models, developed in-house and beyond

Binance uses a combination of proprietary and external technology. Internal models are built to fit the specific risks observed on the platform, while leading external AI and foundation models handle broader reasoning tasks.

This hybrid approach keeps detection at the edge of AI’s capability, while preserving a tailored advantage for Binance’s users, products, and scale. It also means new capabilities can be adopted quickly as the broader AI field advances.

Beyond detection models, Binance maintains an internal Red Team that tests defenses the way a real adversary would— including probing how emerging technologies could be used against the platform.

Jimmy Su, Binance’s Chief Security Officer, explains: “These exercises help us identify weaknesses before attackers do, validate that our controls work in realistic conditions, and continuously strengthen the people, processes, and technology that protect our users. In security, you can’t just assume your defenses will work — you need to challenge them.”

Example: layered defenses against social engineering

Some attacks are designed to completely bypass technology and instead target trust. P2P trading is an area where social engineering attempts are especially common.

At Binance, internal computer vision models detect fake payment receipts by analyzing transaction details and subtle image manipulations. Natural language processing models identify scam messages in chat. Traditional machine learning models assess the risk level of each order. In addition, large language models help interpret the message’s intent and capture embedded scam text inside images.

Machines triage, humans decide, models learn

While AI is exceptional at discovering and initial triage, people remain essential for detailed validation and contextual judgment. Detecting liveness and falsified identity documents is a good example. AI can do a fast pre-triage of large volumes and flag suspicious cases, but many edge cases still require human reviewers to label and judge more accurately.

As new attack patterns emerge, AI helps discover and identify them, and these findings are then used to tune the models. The result is a continuous cycle in which AI provides scale and people provide accuracy. Across all anti-fraud controls, AI models make 80% to 90% of real-time risk decisions and support about 45% of human review workflows.

To keep AI accountable at this scale, Binance applies structured model governance across the model lifecycle—from development and validation to deployment and continuous monitoring. Models are tested for bias and accuracy before going into production, and performance is continuously tracked against real-world outcomes.

When a model’s accuracy drifts or false-positive rates rise beyond acceptable thresholds, as identified by the human review process mentioned above, it is automatically flagged for retraining. This ensures that automated decisions remain fair, explainable, and auditable—not just fast.

Beyond loss prevention: compliance and internal efficiency

Beyond protecting our users, our compliance teams use AI-assisted automation tools to scale and standardize processes such as fraud detection in KYC and transaction monitoring execution.

More than 24 AI initiatives have been implemented across user onboarding, triage escalations, and partner due diligence. These models prioritize cases, identify patterns in complex datasets—including on-chain activity and device fingerprints—and route the right issues to human reviewers, minimizing false alerts.

Internally, the effect is similar. Analysts use AI to move faster in analysis, monitoring, and model deployment, while operations teams use it to help with investigations and case details. In both cases, the goal is the same: hand off repetitive work so people can focus on higher-value judgment decisions. This shows up in adoption—Binance’s internal agentic tool now records roughly 72% usage across teams, supported by company-wide training, prompt engineering programs, and structured supervision.

Underpinning all of this is a commitment to privacy by design. Binance’s AI systems operate within a framework that prioritizes privacy, where data minimization, purpose limitation, and safeguards for user rights are built into how AI models are designed and deployed. The goal is straightforward: protect users from financial abuse and harm without compromising their right to data privacy—and do it transparently, so users understand the safeguards are working in the background without their data being used beyond what the task requires.

Final considerations

At Binance, user protection is a priority that we keep investing in. AI provides scale; people provide judgment. As tools for producing deception get cheaper, this combination must keep evolving—meaning retraining models, refining thresholds, and keeping experienced analysts in the cases where they’re needed.

Additional reading

Legal notice: This content is provided to you “as is,” for general information and educational purposes only, without any statement or warranty of any kind. It should not be interpreted as financial advice, nor is it intended to recommend the purchase of any specific product or service. Digital asset prices can be volatile. The value of your investment may go down or up, and you may not get back the value you invested. You are solely responsible for your investment decisions and Binance is not responsible for any losses you may incur. Not financial advice. For more information, see our Terms of Use and Risk Warning.

Attention: Please note that there may be discrepancies between this original content in English and any translated versions (these versions may be generated by AI). Please refer to the original English version for the most accurate information in case of discrepancies.