0x4e87f5504e87…4e87f54d

ConfirmedSecurity632 vB40 sat/vB3 min decode

Coinbase AI Blocked $4.7 Million in Deepfake Musk Scam

Coinbase says its AI agent blocked $4.7 million, or roughly 91% of attempted customer transfers, during a deepfake Elon Musk Tesla giveaway shown to more than 500,000 viewers.

Coinbase Says Its AI Blocked $4.7 Million in Deepfake Elon Musk Scam
WitnessCoinbase Says Its AI Blocked $4.7 Million in Deepfake Elon Musk ScamHans / pixabay

Outputs

  1. Coinbase says its AI agent blocked $4.7 million, or roughly 91% of attempted customer funds.

  2. The published figures imply about $5.2 million in attempted sends, with roughly $460,000 leaving customer accounts.

  3. The fake YouTube livestream ran for about two hours and attracted more than 500,000 viewers.

  4. A Brooklyn man received a 12-year prison sentence in September over a $16 million Coinbase phishing scam.

  5. A breach at Trezor’s shipping partner exposed 27,000 orders and prompted customer scam calls this week.

By Tom Whitfield

$4.7 million in customer funds did not reach scammers after Coinbase’s AI scam-prevention agent blocked roughly 91% of the money customers tried to send in a fake Tesla cryptocurrency giveaway, the exchange said Tuesday in a weekly consumer-protection post.

The published figures imply customers attempted about $5.2 million, with roughly $460,000 still leaving their accounts. Coinbase said the system worked “with no humans in the loop,” but did not disclose how much of the remaining funds reached the scammers. Coinbase also withheld the incident date and the cryptocurrencies involved.

The fraud centered on a YouTube livestream designed to resemble the real broadcast of Tesla’s flying-car demo. It displayed a deepfake of Elon Musk, on-screen wallet addresses and a QR code, then promised to double any cryptocurrency sent to those addresses. More than 500,000 people watched the stream.

Over roughly two hours, hundreds of Coinbase customers tried to send millions of dollars to the displayed addresses, including many long-time users. Each customer logged in and approved the transfer, placing the activity outside controls built around unauthorized account activity.

How did Coinbase’s agent identify the scam?

Coinbase said attackers hacked no one, broke into no Coinbase account, stole no password and phished no credentials. The platform suffered no breach, and customers initiated the payments themselves.

Coinbase described its defense as having three layers:

  • Machine-learning models score outgoing sends in real time for giveaway patterns, brand impersonation and newly created wallet addresses. The design goal is to catch the danger without treating the customer as the attacker.
  • A scam-specific agent takes over when a model flags a transfer. It looks up context, names the scam pattern, explains how giveaway schemes work and asks questions designed to make the customer’s story fall apart.
  • Coinbase scores each agent response for confidence before showing it to the customer. The system uses that score to determine when a case requires more friction.

The agent had never been trained on this con, and its training data contained no deepfake giveaways tied to live product launches. Coinbase said it nevertheless recognized the scheme through three features: a promise to multiply funds, the deadline of a live event and a transfer that could not be reversed.

Coinbase also said its real-time systems can detect risk signals but cannot determine intent. That limitation defined this case because customers authorized each payment while relying on a fabricated promotion.

What changes as deepfakes become cheaper?

Coinbase said staging the imitation once required a crew, a studio and weeks of work. The company said the current version requires a livestream slot and a model.

Coinbase is expanding AI review of risky sends to every scam type, not only high-profile schemes. It also uses AI to label new scam cases, allowing its models to learn from attacks within hours rather than weeks.

How does the incident fit broader customer fraud risk?

Coinbase described social engineering as a persistent customer risk. In September, a Brooklyn man received a 12-year prison sentence over a $16 million Coinbase phishing scam.

This week, Trezor customers reported scam calls after a breach at the hardware wallet maker’s shipping partner exposed 27,000 orders. On the same day as the Coinbase blog post, the exchange launched a ChatGPT plugin for trading bitcoin, stocks and derivatives. Every order requires users to approve the transaction.

Coinbase closed the post with a list of things it will never do, including promising to double, triple or multiply customers’ crypto. Its warning was direct: “Nobody legitimate will double your crypto.”

Coinbase disclosed no rollout timetable for the broader review. The next control test is whether its agent can identify lower-profile schemes without obstructing legitimate transfers.

via coinbase.com (Original)

More from Tom Whitfield

Tom Whitfield

Show full bio

News editor covering technology and product news in crypto and web3 industry at Mempool Brief.

1,374 articles