Coinbase said its AI scam-prevention agent blocked about $4.7 million from reaching scammers who ran a deepfake Elon Musk giveaway on a fake Tesla livestream.

Elon Musk, whose likeness appeared in the fake livestream the scammer account used. Photo: Steve Jurvetson via Wikimedia Commons (CC BY 2.0). Source
Coinbase said its artificial intelligence scam-prevention agent kept about $4.7 million of customer funds from reaching scammers who ran a deepfake Elon Musk crypto giveaway on a fake Tesla livestream. The exchange described the incident in a consumer protection blog post published Tuesday, Unchained reported on Nov. 18.
The agent stopped about 91 percent of the money customers tried to send "with no humans in the loop," Coinbase said, according to Bitcoin Magazine. That figure and the $4.7 million total match across both reports.
The fake stream ran during Tesla's "flying car" demo. Scammers set up a YouTube livestream that looked like the real broadcast, with a synthetic Musk, on-screen wallet addresses and a QR code, and promised any crypto sent would come back doubled. More than 500,000 people watched, and over about two hours hundreds of Coinbase customers tried to send millions of dollars to the addresses on screen, Coinbase said. Many were long-time users of the exchange. Coinbase did not say when the fake stream ran or which coins customers tried to send.
Unchained's math on the two published figures puts the total customers attempted at roughly $5.2 million and the amount that still left their accounts at about $460,000. Coinbase did not disclose how much reached the scammers; those two derived numbers are Unchained's own calculation rather than figures the company published.
The company stressed that its platform was not breached. "Nobody was hacked," Coinbase wrote. "No Coinbase account was broken into, no password stolen, no credential phished." Customers logged in and approved the transfers themselves, which the exchange said is what makes this fraud hard to catch: its systems "can see risk signals in real time" but "can't see intent," Bitcoin Magazine reported.
The defense has three parts, per Bitcoin Magazine. Machine learning models score outgoing sends in real time, watching for giveaway patterns, brand impersonation and newly created wallet addresses. The goal, the company wrote, is "to catch the danger, not the customer." When a send is flagged, a scam-specific agent takes over, names the scam pattern, explains how giveaway schemes work and asks questions meant to make the story fall apart. Each reply is scored for confidence before the customer sees it, so the system knows when a situation needs more friction, the report said. The flagging models themselves run in the background on every outgoing transfer, watching for signs such as brand impersonation and freshly created wallet addresses.
Coinbase said the agent had never been trained on this kind of con, since deepfake giveaways tied to a live product launch were not in its training data. "It recognized the shape of the scheme anyway," the company wrote, pointing to the promise to multiply funds, the deadline of a live event and a transfer that could not be reversed.
Deepfakes have changed the economics of the scam, Coinbase said. Staging something like this once took "a crew, a studio, and weeks," the post read. "Now it takes a livestream slot and a model." The exchange said it is expanding AI review of risky sends to every type of scam and is also using AI to label new scam cases so its models learn from attacks within hours instead of weeks.
Unchained noted that Crypto Briefing had reported on the scam on Nov. 4, drawing on a Coinbase demo video in which the company put the agent's effectiveness at 99 percent; Tuesday's post measures the result in dollars. Coinbase closed the post with a list of things it will never do, including promising to double, triple or multiply crypto that customers send. "Nobody legitimate will double your crypto," the company wrote.