Coinbase says it fixed an AI-generated World Cup alert after it reported a score before kickoff

Coinbase acknowledged it updated its systems after an AI-generated alert incorrectly reported the result of a FIFA World Cup match before it started, according to reporting that cited remarks from company executives. The alert, shared through Coinbase’s prediction market features, circulated to users as a concluded game outcome even though the match had not kicked off, sparking criticism about the reliability of AI-assisted notifications.

The erroneous alert claimed Norway had beaten Brazil 3-2, with Erling Haaland scoring twice. Users who checked the status of the match quickly flagged the problem, noting that the game had not begun at the time the notification was sent, and that Coinbase’s own prediction market page indicated the match was under a weather delay rather than producing an official result.

A number of crypto and finance commentators criticized the incident as an example of an AI system “hallucinating” factual outcomes and pushing them outward as if they were verified. One point raised in coverage was that prediction-market notifications can reach broad audiences quickly, making errors more consequential than an internal model mistake.

Coinbase Chief Executive Brian Armstrong responded soon after the issue surfaced, urging users to report the problem and indicating the company was reviewing what happened, according to the reporting. Later, an executive (Max Branzburg, as described in coverage) said Coinbase had fixed the incorrect story and made updates intended to prevent similar inaccuracies.

Coinbase’s response, as described across the coverage, frames the incident as a system error rather than a deliberate product change or a change to underlying market mechanics. The company did not, in the cited reporting, disclose specific technical details such as what data source or model step produced the incorrect outcome, or what safeguards failed to stop the alert from going out before kickoff.

For Coinbase, prediction markets are a high-visibility feature because they connect events and outcomes to tradable mechanisms, and they can also support “insight” experiences that summarize what the market appears to be pricing in. Adding AI to that workflow may improve speed and usability, but it also creates a new surface area where incorrect text generation, mismatched event timing, or bad inputs can turn into user-facing misinformation.

The episode arrives as companies across financial services have been embedding AI features into customer workflows, including alerts and automated explanations. The core risk highlighted here is timing and validation, specifically ensuring that any AI-produced event claims are grounded in confirmed, event-complete facts, not in partial, delayed, or ambiguous status indicates.

What Coinbase does not appear to have disclosed in the cited reporting is the scope of the affected alerting system, whether it was isolated to a single match, or what governance and monitoring changes it will implement beyond a general statement that it updated systems to avoid similar inaccuracies. Investors and users will likely look for more transparency on validation, incident procedures, and performance metrics as this becomes a reputational test for AI-powered tooling.