How Binance Says AI Blocked $10.53B in Crypto Fraud

The distinction between blocked potential losses and recovered assets is crucial. A prevention metric estimates risky activity stopped before completion. A recovery metric refers to funds already stolen and later returned or secured. Binance reports both kinds of numbers, but they measure different outcomes.

How Binance says the AI defense stack works

Binance Research says the exchange has set up more than 24 AI initiatives across compliance, with more than 100 AI models powering anti-fraud controls. It also says this anti-fraud stack reduced illicit fund exposure by 96%.

At the center of the description is a custom risk and fraud detection tool called Strategy Factory. Binance Research says it combines business-aware optimization, modular rule construction and continuous refinement, suggesting a system that can adjust detection rules and model-driven risk controls as scam behavior changes.

In practical terms, the stack can be understood in three layers:

1. Detection

Binance says its AI-based risk controls are used to identify threats including deepfakes, phishing scams and AI-powered social engineering attacks. The company’s broader claim is that models and rules work together to spot suspicious behavior earlier, before a risky action turns into a completed loss.

2. Intervention

Report summaries say Binance sends real-time risk alerts at an average pace of more than 9,600 per day. These alerts are the human-facing part of the system: even when models detect risk, users may still need to pause, verify a counterparty or abandon a suspicious transaction.

3. Suppression

Binance also says it has blacklisted more than 36,000 malicious addresses. Address blacklists help exchanges flag or block known scam infrastructure, especially when the same wallets or networks appear across multiple fraud attempts.

Why Binance frames this as AI vs. AI

The exchange’s security narrative is built around a simple problem: attackers are using AI too. Reports on Binance’s latest security push describe crypto security as an “AI vs. AI” arms race because artificial intelligence lowers the barrier for criminals to create deepfakes, voice clones and more convincing phishing campaigns.

Binance Research has warned that AI is currently twice as effective at exploitation as detection in crypto, highlighting an asymmetry between attackers and defenders. Chainalysis’ broader crypto-crime data points in the same direction: it estimated that US$17 billion was stolen in crypto scams and fraud in 2025, while impersonation scams rose 1,400% year over year and AI-enabled scams were 4.5 times more profitable than traditional scams.

That helps explain why AI spending is becoming a financial-crime priority beyond crypto. Binance Research says 75% of financial institutions plan to increase AI spending on financial-crime detection, and it compares Binance’s claimed US$10.53 billion in blocked potential losses with JPMorgan AI systems preventing an estimated US$1.5 billion in fraud losses.

Recovery is separate from prevention

Binance’s largest number is about activity it says was stopped before losses occurred. Its recovery figures are much smaller, and they come from separate anti-scam reporting.

In a 2025 anti-scam update, Binance said it prevented US$6.69 billion in potential fraud and scam losses, including US$3.9 billion related to scam attempts, and recovered more than US$12.8 million in stolen assets. Fortune India separately reported that Binance processed more than 71,000 law-enforcement requests, supported the confiscation of about US$131 million linked to illicit activity and delivered more than 160 law-enforcement training sessions in 2025.

Those categories should not be added together as if they measure the same thing. Potential losses blocked, funds protected, stolen assets recovered and assets confiscated with law enforcement all depend on different definitions and workflows.

What the numbers prove—and what they do not

The figures support one clear conclusion: Binance is positioning AI as core security infrastructure across fraud detection, compliance controls and user protection. If the company’s figures are directionally accurate, its automated defenses are operating at very large scale across millions of users and billions of dollars in risky or attempted activity.

But the numbers still require caution. Most public accounts of the US$10.53 billion figure trace back to Binance’s own disclosure or Binance Research, so the total remains a company-reported prevention estimate. In a separate 2025 context, ICIJ reported that Chainalysis said a Binance report on improving financial crime did not include key crime data, underscoring why exchange crime statistics depend heavily on scope, definitions and methodology.

For users, the practical takeaway is not that AI makes crypto safe by default. It is that exchanges are trying to catch suspicious behavior earlier while scammers use AI to make impersonation, phishing and social engineering more convincing. Risk alerts and address blacklists can help, but the human step still matters: urgent support messages, deepfake videos, pressure to move funds and unfamiliar wallet instructions should be treated as high-risk signals.