0x02f2205d02f2…02f2205a

ConfirmedSecurity473 vB137 sat/vB2 min decode

Coinbase AI Stopped $4.7 Million in Musk Deepfake Giveaway Scam

Coinbase's machine-learning defenses blocked about $4.7 million, or 91%, of transfers attempted during a fake Musk deepfake livestream watched by over 500,000 people.

Coinbase AI Blocked $4.7 Million in Musk Deepfake Scam Transfers
WitnessCoinbase AI Blocked $4.7 Million in Musk Deepfake Scam TransfersPexels / pixabay

Outputs

  1. Coinbase AI blocked roughly $4.7 million, about 91% of attempted transfers, during the scam.

  2. More than 500,000 people watched the fake Musk flying-car livestream over two hours.

  3. No accounts were breached; customers authorized the transfers themselves, many of them longtime users.

  4. Coinbase plans to label new scam cases so models learn within hours instead of weeks.

Coinbase blocked approximately $4.7 million in customer transfers during a two-hour deepfake giveaway scam that impersonated Elon Musk, stopping roughly 91% of the funds customers attempted to send. More than 500,000 people watched the fraudulent livestream, which aired what appeared to be a Tesla flying-car demonstration featuring an AI-generated version of Musk.

The broadcast displayed wallet addresses and a QR code alongside a promise to double any cryptocurrency sent. Hundreds of Coinbase customers tried to transfer funds to the addresses shown on screen during the event.

What distinguished this scam from typical fraud?

No Coinbase accounts were breached. No passwords or login credentials were stolen. Customers signed into their own accounts and approved the transfers themselves, and many of them were longtime users.

That detail matters operationally: because the customers authorized the payments, the activity differed from the unauthorized transactions that fraud systems typically flag. Standard anti-fraud controls, built around detecting account takeover and stolen credentials, do not naturally catch payments that legitimate account holders initiate themselves. The scam exploited that gap by persuading users to authorize transfers voluntarily.

How did Coinbase's AI respond in real time?

Coinbase's machine-learning models scored the transfers in real time, screening for patterns such as giveaways, brand impersonation and newly created wallet addresses. When the system flagged a transfer, an AI agent engaged the customer directly, explained the suspected scam and asked questions before the payment could proceed.

The system also scored the agent's own replies to help decide when to insert an additional verification step before a transfer went through. According to Coinbase, the agent had not previously encountered a deepfake giveaway tied to a product-launch livestream. It nonetheless recognized several warning patterns:

  • A promise to multiply funds sent to an address
  • Pressure to act before the livestream ended
  • Transfers that could not be reversed once completed

Those signals allowed the agent to handle flagged transfers from the first attempt through the last, Coinbase said.

What comes next for Coinbase's risk operations?

Coinbase is investing further in AI for its risk operations. Its stated plans include:

  • Using AI research to update detection models
  • Deploying agents to review high-risk transfers across scam types
  • Labeling new cases so models can learn from attacks within hours instead of weeks

The last point addresses the core operational weakness the scam exposed: detection models trained on labeled historical cases typically take weeks to absorb new attack patterns. A labeling pipeline that feeds fresh cases into models within hours shortens the window in which copycat deepfake campaigns can operate against the exchange's customer base. As AI-generated impersonation of public figures becomes cheaper to produce, real-time behavioral scoring of authorized payments is likely to become a standard layer in exchange fraud defense rather than an experimental one.

via tokenpost.com (Source)

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