Last week, an incident occurred that clearly demonstrates both the potential and the risks of integrating artificial intelligence into financial and information systems. A Coinbase user under the nickname jay_drainjr received a push notification from the exchange, stating that the Norwegian national team had beaten Brazil 3:2 in a FIFA World Cup playoff match, with Erling Haaland scoring a brace. The only problem was that the game hadn't even started yet.
This case is a classic example of an AI hallucination, where the model generates a convincing but completely false result. The notification was generated in the platform's "Trends" section, which uses artificial intelligence to create news summaries based on data from Kalshi prediction markets. Recall that Coinbase recently integrated Kalshi into its ecosystem as part of the "Everything Exchange" strategy — creating a universal trading platform.
Coinbase CEO Brian Armstrong promptly responded to the situation, promising to look into it personally. The next day, the head of consumer and business products, Max Branzburg, confirmed that the error had been fixed and systems updated to prevent similar incidents. Interestingly, the actual match result turned out differently: Norway won 2:1, but Haaland did indeed score two goals. "AI might have known something we didn't," Branzburg joked, although in reality it was just a coincidence.
It is important to note that this incident occurs against the backdrop of Coinbase's active strategy to implement AI. In May, the exchange announced a 14% staff reduction, citing the need to optimize costs and refocus on artificial intelligence. In June, the Coinbase for Agents service was launched, allowing digital assistants to trade on behalf of users. Additionally, the company is participating in Amazon's initiative to introduce payments for AI agents.
My comment: This case is not just a funny curiosity, but a serious wake-up call for the entire industry. When an exchange managing billions of dollars in client funds begins to rely on AI to generate "news," the cost of an error could be catastrophic. Imagine if such a notification triggered mass buying or selling on prediction markets. Coinbase needs not just to "fix the error," but to fundamentally rethink the security architecture of its AI systems, implementing multi-level data verification before sending information to users. Trust is the most fragile asset in cryptocurrencies, and such incidents undermine it far more than it might seem at first glance.