Coinbase’s AI-generated news system sent users a false alert on Sunday claiming Norway had defeated Brazil 3-2 in a World Cup knockout match at MetLife Stadium, with Erling Haaland scoring twice. The problem: the match had not yet started, and Coinbase’s own prediction market page still showed the game as delayed.
What Happened
Screenshots of the erroneous notification spread quickly on X, drawing immediate attention. Coinbase CEO Brian Armstrong responded publicly, saying he was looking into the issue. Max Branzburg, Coinbase’s head of consumer and business products, later confirmed the company had fixed the error and updated its systems.
‘We fixed the incorrect story and made some updates to avoid these types of inaccuracies in the future. It’s awesome to see the power of AI-enabled 24/7 insights for trading, but obviously still need to tune it to address these types of issues.’
Branzburg added a note of accidental prescience: ‘And hey, it turns out Norway did win and Haaland did score 2 goals, so maybe the AI knew something we didn’t!’
The Actual Result
Norway ultimately defeated Brazil 2-1 in East Rutherford, New Jersey, with Haaland scoring twice in the second half. The AI’s fabricated scoreline of 3-2 was wrong on the margin, though directionally correct on the winner and the scorer.
Coinbase declined to comment beyond Branzburg’s public statement on X.
Coinbase’s AI and Prediction Market Push
The incident arrives as Coinbase has been aggressively expanding beyond basic crypto trading. In January, the exchange rolled out prediction markets through Kalshi, allowing users to trade event contracts tied to sports, elections, economic data, and other real-world outcomes. The company has also rolled out AI-powered tools to automate parts of the crypto experience, including AI-powered wallets and agents that autonomously interact with blockchain applications, all part of a broader push to deliver 24/7 market-relevant insights to traders.
The hallucination highlights a known risk with large language model systems used in live publishing contexts: the models can generate plausible-sounding but entirely fabricated outputs, a failure mode that becomes especially consequential when the content touches financial instruments or live event markets where users may act on incorrect data.
