Tesla has put robotaxis into service in limited areas of Austin. The two-seat vehicle has neither steering wheel nor pedals, and forty-five units are reportedly registered in Texas. A federal safety investigation was opened almost immediately. The gap between deployment and regulatory response is measured in hours.
On Wednesday we covered a plane piloted entirely by AI, noting that the debate was moving beyond the military domain. Here is its civilian version, at curb height.
What the absence of a steering wheel changes
Autonomous cars have been operating for years in several American cities. What sets this deployment apart is a design choice: there is no longer any possibility of taking back control.
A vehicle with a steering wheel keeps an exit door. In case of doubt, a human can intervene, even if in practice such intervention is rare and often too late. A vehicle without a steering wheel removes that door. The system decides alone, continuously, with no alternative.
This is the material translation of what we described regarding meaningful human control: supervision that no longer has any means to act ceases to be supervision. Here, it no longer even exists formally.
The sequence, and what it reveals
The most interesting point is not the vehicle, it is the timeline. A deployment in real-world conditions, then an investigation opened on the back of it.
This is the pattern we see in just about every case: the capability arrives, it is deployed, and the framework follows. We saw it with trading opened to agents, with agents that pay, with courts facing AI-assisted cases.
But transport has a particularity: it has a well-equipped regulator and a track record. The automobile has been regulated for a century. The reaction in hours rather than months shows what an existing institutional framework produces when faced with a new capability: it does not prevent it, it examines it immediately.
In our article on trust we described the four conditions that make it rational without understanding: someone understands, a track record exists, failures are analysed publicly, someone answers for them. The automobile has met all four for a long time. A robotaxi accident will be subject to a published investigation, whose findings will change the rules, with an identified manufacturer accountable for it. That is exactly what is missing almost everywhere else in AI, and that is why this sector is a good laboratory for observing how an autonomous technology can integrate into a framework that works.
The question that will arise
A federal investigation does not mean a problem has been found. It means the administration is examining whether the deployment complies with existing rules, and whether those rules are sufficient.
The central question will likely be this: automotive safety standards were written for vehicles with a driver. A vehicle without a steering wheel cannot comply with some of them, not because it is dangerous, but because they presuppose a component it does not have.
This is a classic case of rules written for a world that no longer quite exists. The regulator will have to either grant a waiver or rewrite them, and both options take time.
What to take away
This deployment is a real-world test, and not just for the manufacturer.
It tests the ability of an old regulatory framework to absorb a technology that contradicts its assumptions. It tests what the public accepts when taking back control is no longer possible. And it tests what we have called the delegation criterion: reversible, explicit criterion, assessable outcome.
A trip in the city is reversible in most cases, the criterion is clear, and the outcome can be observed. This is probably one of the areas where full autonomy is most defensible. What remains to be established is the real error rate compared with that of humans, over a sufficient volume. Forty-five vehicles will not tell us that. Forty-five thousand will, and that is what is at stake.