Coinbase said its AI scam defenses blocked about $4.7 million in attempted transfers to a fake crypto giveaway promoted by an Elon Musk deepfake. According to Coinbase, its automated systems blocked around 91% of the attempted transactions before any funds could be delivered, without requiring intervention from employees. Hundreds of Coinbase customers tried to send millions of dollars to wallet addresses shown in the fake livestream within roughly two hours.

How the scam worked

The scam used an AI-generated deepfake of Elon Musk during Tesla's flying car demonstration and attracted more than 500,000 viewers on YouTube. Scammers timed the fraudulent livestream around the demonstration, and the YouTube broadcast imitated Tesla's own. The digital imitation of Musk urged viewers to send cryptocurrency, with wallet addresses and a QR code shown on screen. The broadcast promised to double any cryptocurrency viewers transferred. The attackers used the Musk deepfake to lend credibility to the fake offer and drive engagement among cryptocurrency users.

Customers approved the transfers themselves

Coinbase said attackers did not break into affected customers' accounts or steal their passwords, and that customers logged in and approved the transfers themselves. Customers authorized the transfers after being deceived by the fake giveaway, so the transactions appeared legitimate.

How Coinbase's system responded

Coinbase said its machine learning models assessed outgoing transfers for warning signs, including giveaway patterns, brand impersonation and newly created wallets. When a transfer triggered concern, a dedicated scam prevention agent examined its context, explained the scheme, and asked targeted questions meant to challenge the customer's assumptions. Rather than showing a generic fraud warning, the agent started a targeted conversation explaining the specific giveaway pattern. Coinbase said the intervention operated without humans in the loop, with the system identifying malicious activity and engaging affected customers automatically. Coinbase said the agent recognized the scam despite lacking training on this specific deepfake giveaway format. It said the agent had not seen this format but recognized warning signs including promises to multiply crypto, time pressure and irreversible transfers. Coinbase urged users to verify promotions through official company channels before transferring cryptocurrency.

Comparison with a 2020 scam

A comparable scam took place in July 2020, when several prominent Twitter accounts, including Elon Musk's, were compromised and an offer claimed funds sent to a listed address would be doubled. Coinbase then responded by manually blacklisting the scammers' Bitcoin address, but only after payments were already underway. In the current incident, its automated system detected users' transfer attempts and blocked them before funds were drained.

Wider crypto theft figures

DefiLlama data show 343 major crypto attacks in 2026 so far, with $8.78 billion in stolen assets. TRM Labs said in August that reported losses from deepfake scams in 2026 had already increased 263 percent compared with the full-year total for 2025. Investigative authorities identified the North Korean hacking group TraderTraitor as behind what they called the largest cryptocurrency theft in history, linked to a Binance-related hacking case in October.