Binance's artificial intelligence security system intercepted $10.53 billion in scams over a 15-month period that began in 2025, the exchange revealed. The AI tools flagged and blocked 36,000 malicious addresses, preventing the loss of user funds from phishing, rug pulls and other fraud.
The $10.53 billion scam net
The figure covers scams detected between early 2025 and mid-2026. That's roughly $700 million a month the system stopped before it could hit users' wallets. The 36,000 blocked addresses represent the digital infrastructure scammers used to drain accounts or trick victims into sending crypto.
Binance didn't break down the total by scam type, but the numbers underline how aggressive fraud has become in crypto. The AI appears to catch both high-value targeted attacks and smaller, mass-distributed schemes.
How the AI identifies threats
Binance's security layer uses machine learning models trained on transaction patterns, wallet histories and known scam indicators. When an address matches suspicious behavior — like a sudden burst of small test transactions followed by large withdrawal requests — the system blocks it automatically.
The 36,000 addresses is not a small list. Each one could have been used in multiple scams over time, so blocking them early cuts off entire fraud operations. The AI also adapts as scammers tweak their methods, a constant game of cat-and-mouse.
What the money saved means
Ten and a half billion dollars is real money, even for a platform that moves billions daily. For individual users, the AI likely saved thousands from losing their savings. The exchange hasn't said how many users were directly protected, but the address-level blocking suggests broad coverage.
The 15-month window shows the persistence of crypto crime. Despite the industry's efforts, scammers keep trying. Binance's AI is one layer — but no system is perfect. The company continues to refine its detection as new schemes emerge.
The fight isn't over. With each new block, scammers look for the next gap. Binance's AI team stays on watch.
