News › AI
AI
Deepfakes are everywhere now: Meta’s AI chief posts one of Sam Altman, and Coinbase’s AI stops $4.7 million going to a fake Musk
Two reports from 17 November show both sides of the same technology: a labelled deepfake used as marketing, and a scam where hundreds of customers tried to send money to a fake Tesla giveaway.


Two stories published on 17 November show how ordinary synthetic video has become. In the first, The Verge reports that Alexandr Wang, Meta’s chief AI officer, shared a short AI-generated clip earlier this month to promote Muse, the company’s consumer AI agent. In it, Wang, Mark Zuckerberg, OpenAI chief executive Sam Altman and Muse’s mascot Jolly are on a four-way video call, and Altman joins dressed in a furry suit and face paint to look like Jolly. The caption reads “When the homies meet about openai product strategy.”
The Verge notes that a few years ago such content would have drawn strong public rebukes and threats of legal action. Meta requires AI-generated videos depicting people to be labelled as AI, and Wang’s post is labelled, a spokesperson said; static images carry no such requirement. The spokesperson also pointed out that Altman has made his likeness available for AI content generation, as when OpenAI launched its Sora app. OpenAI did not respond to requests for comment. The article adds that political groups ran millions of dollars of AI-generated ads before the midterm elections, and that a fake image of actress Mia Goth spread in October from a verified fan account, which her representative called “incomprehensible.”
The second story shows the harmful side. The Block reports that during Tesla’s “flying car” demonstration in October, scammers ran a YouTube livestream that looked like Tesla’s official broadcast, featuring a deepfake of Elon Musk promoting a fake giveaway that offered to double any cryptocurrency sent. It showed wallet addresses and a QR code. According to Coinbase, more than 500,000 people watched, and within about two hours hundreds of Coinbase customers tried to send millions of dollars to the addresses.
Coinbase says its AI defenses stopped $4.7 million, about 91% of the attempted transfers, without human involvement. Machine-learning models scored risky sends in real time, flagging giveaway patterns, brand impersonation and newly created wallet addresses. A dedicated scam-prevention agent then reviewed the transaction context, explained how the scam was structured and questioned the customer’s story. Coinbase said the agent had never seen this exact scam before, but recognised the pattern: a promise to multiply funds, urgency from an event ending within hours, and a transfer that cannot be reversed.
Coinbase stressed that no accounts were hacked and no passwords or credentials were stolen: customers sent money voluntarily, and many were long-time users. That is the hard part of the problem, as The Block notes: most platforms use real-time fraud detection, but stopping transfers that people were tricked into authorising is difficult. Coinbase said it will keep investing heavily in AI defense.
Together the stories give a simple rule. A video of a famous face is no longer evidence of anything. Never send funds to an address shown in a livestream, nobody legitimate doubles your coins, and a transfer on a public network cannot be undone, so the pause before sending is your last line of defense.
Facts only, never financial advice.