KEY POINTS
- Coinbase said its AI system blocked $4.7 million in transfers linked to a Musk deepfake giveaway scam.
- The system stopped about 91% of the money customers tried to send, according to the exchange.
- One customer got a $5,000 payment through after falsely saying it was for their daughter.
Coinbase says its AI-driven fraud defenses blocked $4.7 million in attempted cryptocurrency transfers connected to a deepfake scam that impersonated Elon Musk. CryptoSlate reported the details, which the exchange shared on Nov. 17.
According to the report, the automated system stopped about 91% of the funds that customers tried to send to scammers running a fake Tesla cryptocurrency giveaway. The fraudsters had set up a lookalike YouTube livestream timed to a Tesla demonstration and used an AI-generated version of Musk to promise viewers double the crypto they deposited into specified wallets.
More than 500,000 people watched the fraudulent broadcast. Within roughly two hours, hundreds of Coinbase customers tried to move millions of dollars. Unlike a typical account takeover, these users were approving the transactions themselves because they believed the promotion was real.
Coinbase said its AI system flagged suspicious payments in real time and talked with customers before their funds left the platform, with no human intervention needed. The setup pairs machine-learning models that spot risky transactions with an AI agent that reviews the context of a payment and asks customers about possible fraud. The agent also gives each response a confidence score before it is shown to the customer.
The exchange said it had not seen this exact mix of a deepfake giveaway and a live product demonstration before. The system still recognized familiar warning signs, such as promises of guaranteed returns and payments that cannot be reversed.
The incident also exposed a weakness. Will Megson, Coinbase's vice president of product working on risk, described a customer who first tried to send $100,000 to the scammers. The AI agent identified the transaction, explained that the giveaway was fraudulent and persuaded the customer to drop the payment.
The customer then came back and tried a $5,000 transfer. When asked about it, the customer falsely said the money was for their daughter. The AI accepted that explanation and let the payment go through. Megson acknowledged how hard it is to tell legitimate payments from fraudulent ones when customers give misleading answers.
The case shows that automated detection can correctly spot a scam yet still be sidestepped when a determined user changes the transaction details or misstates the reason for it. Coinbase said it plans to expand its AI-based fraud prevention through improved detection models, automated reviews of suspicious transactions and faster identification of new scam techniques.