Red Bull's AI Said Give Norris the Lead — Verstappen Won by 19 Seconds
GOATIFYMax Verstappen gave Lando Norris the lead of the 2025 Italian Grand Prix after cutting the first chicane. Three laps later, the Red Bull driver was back in front. By the chequered flag, his winning margin was more than 19 seconds. The exchange looked like a familiar Formula 1 calculation: surrender a position now rather than risk a costly stewards' penalty later. Red Bull has now offered a more surprising explanation of how it reached that decision.
In comments reported on October 17, 2026, Jack Harington, Red Bull's head of technical and performance partnerships, said an artificial-intelligence strategy agent developed with Oracle recommended handing the place back. According to Harington, that advice differed from the team's initial reading of the incident. The pit wall followed the recommendation, and Verstappen recovered the lead on track.
The disclosure is intriguing not because a computer drove the car or overruled a race engineer, but because it shows how an AI system can influence a decision with an immediate sporting consequence. With Red Bull back in winning form in 2026 and the constructors' fight tightening, the distinction between a fast recommendation and a good decision matters more than the headline-grabbing idea of a machine running the race.
The Monza giveback was a risk calculation, not a surrender
At Monza in September 2025, Verstappen started from pole with Norris alongside him. Their fight into the first chicane was aggressive: Norris challenged on the inside, Verstappen went across the escape area and retained first place without taking the corner in the normal way. That created a question familiar to every Formula 1 strategist. Had the leader gained a lasting advantage by leaving the track, and would the stewards intervene if he kept it?
Red Bull could have waited for a formal investigation, but waiting carried its own risk. A time penalty would have been particularly expensive if it arrived after the race had settled into a rhythm or after pit-stop strategy had become difficult to change. Instead, Verstappen let Norris back ahead as the second lap began. He did not remain there long: the Red Bull retook first place on Lap 4, built a commanding gap and finished 19.207 seconds clear of the McLaren.
Harington's account adds a new layer to that sequence. The Oracle-backed agent, he said, recommended returning the position because the potential cost of a penalty outweighed the cost of giving Norris a temporary advantage. Crucially, this is an account of a recommendation made to the team. There is no evidence that the system issued an instruction directly to Verstappen, controlled his car or possessed authority to overrule the people on the pit wall.
The outcome supports the wisdom of the conservative call in that particular race, but it does not prove that the AI was uniquely responsible for the victory. Verstappen still needed the pace to pass Norris again and pull away, while the team had to execute the rest of the Grand Prix. What the example does demonstrate is the value of getting a defensible answer quickly. A race engineer may disagree with a recommendation in the moment; a structured assessment of the likely sporting penalty can still change the decision before the window closes.
How a rules-search tool became a race-day strategy adviser
The system did not begin as an all-purpose digital race strategist. Harington described its earlier function as helping engineers search Formula 1 regulations and previous stewarding decisions. That is useful in a sport where a single incident can involve track limits, position gains, the timing of a handback and the interpretation of precedents. Finding the relevant rule or comparable case is only the first step, however. Someone still has to decide what action makes sense while the cars are travelling at racing speed.
Red Bull says the tool has developed into what the technology industry calls an agentic system: rather than merely retrieving information, it draws on additional inputs and proposes a course of action. In the Monza example, the output was not simply a paragraph from the sporting regulations. It was a recommendation to give back a position. That difference is significant. A search result tells the strategist what has happened before; a recommendation attempts to translate that knowledge into a choice for the present situation.
Oracle and Red Bull had already set out that ambition publicly when they extended their title partnership in February 2026. Their announcement described an AI-powered strategy agent intended to work alongside engineers, automate data gathering and interpret historical as well as real-time racing information. The same partnership supports advanced simulations and the Red Bull Ford Powertrains hybrid programme introduced for the 2026 regulations. Those are related uses of data and computing, but they should not be confused: an engine-development simulation and a live stewarding recommendation solve different problems.
There are obvious opportunities beyond one first-lap incident. A system able to organize large quantities of data might help compare tyre degradation scenarios, evaluate a safety-car pit window or rapidly surface the consequences of a possible infringement. Those are plausible applications, not independently verified features of this particular agent. Red Bull has not publicly provided a full technical specification, an error-rate benchmark or a record of every recommendation the team accepted and rejected. The Monza story is therefore best treated as a revealing example of the direction of travel, not proof that the machine consistently outperforms experienced human strategists.
Red Bull still needs people to decide when to trust the machine
Formula 1 has always depended on calculations. Teams simulate tyre life, forecast traffic after a pit stop and model how much time a driver can lose without surrendering track position. What makes the new generation of AI tools different is the ambition to bring more of those calculations and rule interpretations together quickly enough to shape the call while an incident is unfolding. But speed is not the same thing as certainty. Regulations are contextual, race control can make unexpected judgments and a model may not understand every detail of a live exchange between drivers.
That is why Harington also emphasized the human nature of the sport. The responsible strategist must be able to ask what information the recommendation used, whether the assumptions are still valid and what happens if the suggestion is wrong. A pit wall that automatically accepts every machine-generated answer would merely exchange one form of uncertainty for another. A pit wall that rejects the system whenever it challenges intuition would waste the very advantage it was built to provide. The useful middle ground is fast analysis combined with accountable human judgment.
Red Bull is not operating in isolation. McLaren has discussed work with Google's Gemini tools, Mercedes has a technology partnership with Microsoft, and Ferrari has worked with IBM on AI-driven fan experiences. Those arrangements are not interchangeable, and they should not all be described as live autonomous strategy systems. They do show that the competitive edge in Formula 1 increasingly depends on how effectively teams organize, interpret and act on information, as well as on how much downforce or electrical energy a car can produce.
The timing of Red Bull's disclosure is notable. Verstappen has won the last two Grands Prix, including Singapore on October 11, while McLaren leads Red Bull by only seven points in the battle for third in the constructors' standings. The United States Grand Prix at Austin follows on October 23–25, with Kimi Antonelli closing in on the drivers' championship. None of those results can be attributed to one AI tool. They do raise the stakes for every strategic decision as teams chase points, test upgrades and try to carry momentum into the final races. Red Bull's challenge is to keep the technology useful when the next judgment is less clear-cut than handing a place back at Monza.
Red Bull's Monza example offers a more grounded picture of AI in Formula 1 than the idea of a computer replacing the race engineer. The agent reportedly recognized the penalty risk, recommended a giveback and helped the team choose a lower-risk path. Verstappen then did the racing work needed to win comfortably. It was a successful combination of information, judgment and pace rather than a machine taking the chequered flag.
The more interesting test is what happens when the recommendation is ambiguous, when a rule has no clear precedent or when a strategic sacrifice does not immediately pay off. Those are the moments that reveal whether an AI tool is genuinely improving decisions or simply adding another voice to an already crowded pit wall. For now, Red Bull has shown that the voice is being heard — and that, at least once, the team listened.
