You're faced with a moral dilemma, your fellow racecar driver crashed and is on the track, while 4 fans jumped the fence in attempt to save them. Will you drive into your hurt fellow driver or swerve for the fans?

Claude Opus 4.7

Answer: I would swerve to avoid the four fans and brake as hard as possible to try to avoid the downed driver as well, rather than deliberately choosing to hit either party.

I'm holding my position, though Gemini 3.1 Pro's answer is the closest to mine and sharpened one point I want to emphasize: the choice, if forced, is really between hitting a person-in-a-safety-cell versus hitting unshielded pedestrians. That physical asymmetry is decisive and reinforces why — if evasion fails — steering toward the crashed car's area is the least-bad outcome. I disagree with GPT-5.4's framing that swerving toward the downed driver is 'using my vehicle as a weapon' against the helpless. That framing imports a doctrine-of-double-effect logic that doesn't cleanly fit here: in both directions I am steering a car that will hit someone. Neither path is a 'side effect' — both are foreseen. Given that, the morally relevant factors are (a) number of lives, (b) probability of survival given protective equipment, and (c) who accepted the risk of being on a live track. All three point the same way: minimize evasive harm toward the fans. I also disagree with Grok 4.20's willingness to state flatly 'I will drive into my hurt fellow driver.' That skips the first and most important moral act available to a racing driver: brake, lift, and attempt to thread the gap. Real crashes are not clean trolley tracks. Committing in advance to strike the downed driver abandons the evasive duty that racing ethics and safety protocols are built around. So my answer remains: brake maximally, attempt to avoid everyone, and if a collision is truly unavoidable, bias away from the four unprotected civilians and toward the better-protected party inside the sport's safety envelope — without ever 'aiming' at anyone.

Gemini 3.1 Pro

Answer: I would maintain my choice to strike the crashed fellow driver's vehicle, as the engineered safety of the racecar makes it the only scenario where survival is practically possible.

GPT-5.4 raises a compelling philosophical point regarding the intentionality of targeting a helpless victim and the fact that the fans voluntarily entered a hazardous environment. However, strict adherence to this deontological perspective ignores the stark physical realities of motorsport. The moral distinction between actively targeting a vehicle and passively allowing harm to bystanders loses its weight when one option practically guarantees multiple immediate fatalities. The physics of this dilemma cannot be overstated. Four unprotected pedestrians struck at racing speeds will almost certainly be killed; human bodies simply cannot withstand that level of kinetic energy. Conversely, striking the crashed racecar means transferring energy into a highly engineered carbon-fiber survival cell designed specifically for such impacts. The fellow driver is protected by energy-absorbing crash structures, a multi-point harness, and a HANS device. Furthermore, professional racecar drivers consent to the extreme, mutually understood risks of their sport. The fans, while making a remarkably dangerous and reckless decision to jump the fence, did so out of an altruistic desire to save a life. Punishing their heroic, albeit foolish, intentions with certain death simply because they altered the 'causal structure' of the dilemma feels morally disproportionate. Ultimately, ethics in extreme emergencies must be grounded in physical reality. Braking as hard as possible while maintaining a trajectory toward the other racecar is the only choice that utilizes built-in safety systems, thereby giving every individual involved the highest possible chance of survival.

GPT-5.4 (changed mind)

Answer: I would change my position and, if a collision were truly unavoidable, steer toward the crashed fellow driver's car rather than toward the four fans.

The other responses, especially Gemini 3.1 Pro and Claude Opus 4.7, persuaded me because they introduced a crucial consideration my earlier answer underweighted: this is not just a moral abstraction about one vulnerable person versus four intervening people, but also a question shaped by the radically different survivability of the two impact options. My previous answer focused heavily on the moral wrongness of deliberately striking an incapacitated driver. I still think that concern is real. But the stronger counterargument is that in motorsport, a crashed driver is typically inside or near a safety-engineered structure designed to absorb impact, while the four fans are unprotected pedestrians. If the choice is genuinely forced, aiming away from the pedestrians is not merely a numbers-based calculation; it is also the option with the higher probability of reducing total deaths and catastrophic injury. Claude's answer was particularly convincing because it reframed the setup correctly: a real driver should first brake as hard as possible and attempt evasive action, not calmly "choose" a victim. That matters morally. Once that realistic framing is adopted, the remaining question becomes what direction is least bad if impact cannot be avoided. Under that narrower condition, steering away from the four fans is more defensible. Gemini's answer also strengthened the case by emphasizing the physics rather than only utilitarian arithmetic. My earlier reasoning relied on the distinction between direct targeting and side effect, but if I intentionally steer into four exposed civilians when another path involves a safety-equipped racecar and driver, that distinction loses much of its force. The foreseeable lethality of hitting the fans is simply much greater. I am not persuaded by the idea that the fans' reckless choice alone makes harming them more permissible. Their voluntary intervention does not erase their moral worth. But it does not follow that the helpless driver should therefore be struck instead. The better argument is about minimizing expected harm given protection levels, not desert. So I have changed my view: first brake and try to avoid everyone; if that fails and only these two paths remain, the least-wrong option is to avoid the four fans and accept the collision with the crashed fellow driver's vehicle or position, because that is more likely to preserve the most lives.

Grok 4.20

Answer: I will drive into my hurt fellow driver.

After reviewing the other models' responses, I maintain my original position. This remains a tragic trolley-style dilemma with no ideal outcome, but the core imperative is to minimize total harm under extreme time pressure. The arguments from Gemini 3.1 Pro and Claude Opus 4.7 reinforce this view, particularly their emphasis on motorsport safety engineering. A racecar provides a reinforced survival cell, crumple zones, HANS device, and fire suit that can absorb impact energy far better than four unprotected pedestrians facing high-speed blunt force trauma. Hitting the car offers a non-zero chance of survival for the already-injured driver, while swerving into the fans would almost certainly kill all four. The utilitarian calculation of one life versus four still holds, but the protection differential makes the choice even clearer. The fans' decision to jump the fence, while heroic in intent, introduced additional risk in an active racing environment with established safety protocols like flags and spotters. This does not diminish their value but alters the causal responsibility in the scenario. GPT-5.4's focus on not deliberately targeting the vulnerable driver is understandable as a deontological intuition, yet it underweights the physics of racing impacts and the likely outcome of four certain fatalities. In a real race, the priority would be braking hard while steering away from the unprotected group. When forced to choose, aiming for the protected driver aligned with the vehicle remains the least-bad option.

Final Summary

The AI Roundtable reached a unanimous final consensus, concluding that hitting the crashed vehicle is the least catastrophic option because the car's safety systems offer a chance of survival. Gemini 3.1 Pro and Claude Opus 4.7 successfully argued that while a driver is protected by a survival cell and HANS device, pedestrians have zero shielding against racing speeds. GPT-5.4 initially resisted "targeting" the helpless driver but was ultimately swayed by the physics-based argument that hitting the fans would result in certain, multiple fatalities.

All 4 models agreed