Coinbase AI sends false breaking news alert about match that hadn't started yet

Coinbase AI Fabricates Match Result Before Kickoff

On July 5, Coinbase AI sent millions of users an AI-generated breaking news alert claiming Norway had beaten Brazil 3-2 in a FIFA World Cup quarterfinal match that hadn't even started 1

1

. The alert came from Trends, an AI-generated feature inside Coinbase's prediction markets app that surfaces news summaries alongside real-time market data 2

2

. At the time the alert was sent, Coinbase's own market page showed the game under a weather delay 2

2

.

The AI hallucination proved particularly concerning given Coinbase's partnership with prediction markets app Kalshi. Users on social media called out the company for the "dangerous and irresponsible" error, noting that a hallucinated news alert could have caused bettors to lose money based on entirely fabricated information 1

1

. This incident underscores the risks of AI hallucinations in real-time applications where financial decisions hang in the balance.

AI Told Millions Norway Won, Then Reality Caught Up

What made the incident stranger was that Norway did eventually beat Brazil, winning 2-1 with Erling Haaland scoring twice in the second half 2

2

. However, the AI had invented the score, the timing, and the certainty, delivering all of it as confirmed news to users with active positions on the outcome 2

2

.

Max Branzburg, Coinbase's head of consumer products, claimed the company had fixed the incorrect story and made updates to avoid these types of inaccuracies in the future 1

1

. Despite the significant error, Branzburg defended the "power of AI-enabled 24/7 insights for trading," while conceding Coinbase still needs to "tune it to address these types of issues" 1

1

. He even suggested "maybe the AI knew something we didn't" since Norway did win and Haaland scored 2 goals 1

1

.

AI Skipping Uncertainty Creates Gambling Chaos Risk

The contrast with how Opta approaches the same tournament captures the difference between calibrated probabilistic forecasting and confident fabrication. Opta's supercomputer has run 25,000 simulations per stage to estimate each team's probability of winning 2

2

. A model that distributes probability across outcomes is a forecasting tool; a model that declares Norway beat Brazil 3-2 before kickoff presents fabrication as fact 2

2

.

Experts warn that combining AI-generated misinformation with the growing obsession over prediction markets like Kalshi and Polymarket creates a fatal mix that could trigger gambling chaos, particularly as concerns mount over surging gambling addiction 1

1

.

Broader Pattern of AI-Generated Misinformation

Coinbase isn't the first to send entirely incorrect alerts. Apple was forced to pull an AI feature that consistently delivered bogus news alerts to iPhone users in January of last year 1

1

. The BBC filed a formal complaint after the AI falsely told users that Luigi Mangione had shot himself, when he remains alive and awaiting his federal trial postponed to January 2027 1

1

.

The issue remains baffling considering how long companies have dealt with this exact problem. Since the dawn of large language model-based tools like ChatGPT, the bots have been confidently fabricating facts despite hundreds of billions of dollars spent on the tech's development 1

1

.

Financial Services Face Higher Stakes

The reliance runs deeper than one bad alert. Coinbase has said roughly 40% of its daily code is now AI-generated, with plans to push that figure higher 2

2

. Sports predictions carry low stakes compared to where the same failure mode appears in financial services. PYMNTS Intelligence found that 60% of financial services firms use AI for credit risk assessment and scoring, with 85% increasing their AI budgets over the next 12 months 2

2

.

Fraud detection, risk assessment, and revenue forecasting are dominant use cases where a confident but fabricated output causes the most damage 2

2

. When a model makes an error in a financial workflow, money moves incorrectly at speed across thousands or millions of transactions before anyone detects the problem 2

2

.