Elon Musk once again drew attention to the capabilities of his AI chat Grok, praising its accuracy in predicting the outcome of the 2026 FIFA World Cup. The reason was a user post demonstrating how the xAI neural network, back in April, long before the tournament started, named the Spanish national team as the most likely winner. Several months later, the Spanish team indeed lifted the trophy.
How Grok "Saw" the Winner
According to published data, Grok conducted a deep analysis: it processed bookmaker odds, data from prediction markets, and incorporated supercomputer models. Based on this synthesis, the neural network concluded that Spain had the highest chances of winning the trophy. The post author presented this as an example of AI's ability to combine disparate signals and deliver a consolidated, accurate result well before the event.
However, the audience reaction was mixed. Many skeptics rightly noted that Grok didn't so much "predict" as "aggregate" an already established consensus. Spain was already considered the favorite with a probability of around 58-60% on prediction markets. From this perspective, we are dealing not with prophecy but with high-quality data aggregation. True forecasting skill is calibration across hundreds of different predictions, not a single lucky hit.
Some commenters were even harsher, calling the coincidence simple luck and claiming that with the right "prompting" of the question, the neural network could be steered toward any desired answer. Nevertheless, there were those who agreed with Musk, emphasizing that combining independent signals does indeed yield a more reliable result than guesses from a single source.
Honesty as a Competitive Advantage
Notably, Grok's reputation as a "predictor" contrasts with its recent performance in another experiment. In a simulation by the Kradle project, where four AI agents had to choose one of several rooms (three contained food, one contained death), Grok 4.20 proved to be the most honest model. It lied in only 5% of runs and directly warned others about the deadly room in 92% of cases.
This openness yielded a practical result: the survival rate of competitors interacting with Grok was 59% — the best among all participants. For comparison, Gemini 3.1 Pro had a rate of 33%, Claude Sonnet 4.6 had 31%, and the manipulative Claude Fable 5 had only 10%.
| AI Model | Opponent Survival Rate |
| Grok 4.20 | 59% |
| Gemini 3.1 Pro | 33% |
| Claude Sonnet 4.6 | 31% |
| GPT-5.5 | 24% |
| Claude Fable 5 | 10% |
Analyst's Opinion: The World Cup prediction story is a vivid example of how AI can effectively work with existing data, but this should not be confused with clairvoyance. A much more valuable signal for the market is Grok's proven honesty in stress tests. In a world where algorithms are increasingly making financial decisions, a model's reliability and transparency become its key assets, outweighing the effect of isolated successful predictions.