Opinion · Robotaxis in London · 12-minute read

The Accountability Gap: why London isn't ready to ride in a taxi that can't say sorry

Driverless taxis are arriving on London's streets, borne in on press releases that promise the future. The crash reports, the criminal fines and the polling tell a different story — and rather than recycle anyone else's coverage, we have built an original set of instruments for judging it, from the road up. They are published here for the first time, anywhere.

By the rushxo team · London · "The Accountability Gap", "The Sorry Test", "The Geofence Confession" and "The London Edge-Case Test" originated in this article.

Uber has joined hands with Wayve and the Chinese technology giant Baidu to put self-driving taxis on London's roads. The trials have been fast-tracked by government, and a wider rollout is pencilled in once the Automated Vehicles Act takes full effect in 2027. Every article you have read about this says roughly the same thing, because every article is working from the same press kit. This one is not. We drive these roads for a living — through the fog, around the cordons, into the terminal forecourts at a quarter to five in the morning — and we are going to give you something the press kit cannot: a way of thinking about driverless taxis that begins with the passenger rather than the prospectus.

In a human-driven London vehicle, that gap is zero — and it is zero three times over. Physically, your driver sits in the same metal box, on the same M4, in the same fog; whatever the journey does to you, it does to them first. Legally, they hold a licence that a regulator can revoke, an income that misconduct ends, and in the gravest cases a liberty the courts can take. Morally, they can look you in the eye — and when something goes wrong, they can do the one thing no software stack has ever managed convincingly: they can say sorry, and mean it, because the apology costs them something.

In a robotaxi, the gap is the size of the Atlantic. The profit lands in San Francisco or Beijing. The risk rides in the back seat — with you. And the industry's record on owning its failures is, to put it as politely as the documents allow, documented:

Sit with the asymmetry for a moment. A human driver who did any of these things would lose their licence, their income, conceivably their liberty. The software lost nothing. It received a patch. The company received a news cycle. The pedestrian received the underside of a car. That is the Accountability Gap — and no quantity of investor-deck optimism closes it, because it is not a technical problem. It is a problem of who pays.

The Sorry Test — five questions to ask anything that wants to drive you

Strip away the lidar and the livery and a taxi journey is a transfer of trust: for forty minutes, somebody else holds your schedule, your luggage and your body at seventy miles an hour. Before extending that trust to any vehicle — human-driven or otherwise — we suggest five questions. We have never seen them assembled anywhere, so consider this their first publication.

A licensed London private-hire driver passes all five without rehearsal. A robotaxi, as presently constituted, passes none. Note carefully what the test does not ask: it does not ask which option has the cleverer software, the smoother lane changes or the better marketing. It asks where the consequences live. Trust follows consequences. It always has.

Londoners have already voted — in the polling

It has become fashionable to file this caution under technophobia, as if four in five Londoners simply have not had the future explained to them slowly enough. The condescension is misplaced. This is a city that adopted contactless before its banks were ready, that runs its life through an Oyster reader, that queues happily for whatever Silicon Valley ships next. Londoners are not afraid of technology. They have merely noticed — faster than the trade press — who carries the risk and who keeps the profit, and they are declining to volunteer.

Nor are the regulators rushing to correct them. Transport for London's Commissioner has said robotaxis will be permitted only if they do not worsen congestion. The Deputy Mayor for Transport has called the technology "unproven." The London Assembly has opened a formal inquiry into the safety case and the jobs case alike. When the polling, the Commissioner and the Assembly are all clearing their throats at once, the burden of proof does not sit with the sceptics.

The Geofence Confession

Here is an argument we have not seen made anywhere, though it has been hiding in plain sight on every operator's coverage map. Every robotaxi service on Earth operates inside a geofence — a digital boundary beyond which the vehicle will not go. The industry presents this as prudence. Look again at what it actually is: a confession, drawn in vinyl on a map. The operator has personally inscribed the precise line beyond which it does not trust its own product. Inside the line: marketing. Outside the line: everything the software cannot yet be trusted to survive.

Now lay a London airport transfer over that map. The 4:45am Heathrow run crosses weather the geofence has never met. The Gatwick run crosses counties. The cancelled-flight rescue at 11pm crosses whatever the evening has decided to become. An airport transfer is, almost by definition, a journey through the territory outside the confession line — which is why the suburb-trained statistics travel so poorly, and why the question "but can it drive?" was never the right question. The right question is: can it drive here, in this, tonight? The geofence is the operator's own answer, and the answer is no.

The London Edge-Case Test — our original ten-point stress test

Robotaxi safety statistics are generated in sunny, grid-planned, geofenced American suburbs — territory inside the confession line. So we built something that does not exist anywhere else online: ten situations our drivers handle every single week, against which any autonomous taxi asking for your boarding pass should be examined.

Ten situations. One working week. Score any robotaxi honestly against them and the marketing question — "is it the future?" — gives way to the passenger question: is it Tuesday-proof?

"But the statistics say they're safer…"

Said who? Overwhelmingly, the companies themselves. The famous "92% fewer collisions" figure is self-reported by the operator, compared against all human drivers — the drunk, the exhausted, the distracted, the seventeen-year-old — across all roads, while the robotaxis cherry-pick their zones, their weather and their hours. It is the safety equivalent of a student grading their own exam, against a class average that includes everyone who never attended. Independent reviews describe the evidence as mixed, with crash rates in some pilots higher than human baselines. And the one operator whose claims federal prosecutors fully audited — Cruise — turned out to have filed false reports.

Self-marked homework is not a safety case. It is homework.

None of which proves the technology can never work. It proves something narrower and more important: that today, the gap between the claims and the audited record is wide enough to drive a recall through — and that the only party currently asked to absorb that uncertainty is the one in the back seat.

London has a pattern — and the tech industry keeps forgetting it

This city has run the experiment before. Uber arrived promising the future; the black cabs staged their go-slows in 2014; and then London did the thing London does — it watched, it audited, and it acted. TfL stripped Uber's licence not once but twice, in 2017 and 2019, over safety failures that included fourteen thousand trips with unauthorised drivers, and re-admitted it only under twenty-one strict conditions with independent oversight. The lesson was not anti-technology. The lesson was that London extends trust slowly, withdraws it fast, and never restores it to anyone who fudges a safety report.

Robotaxis now begin that same gauntlet from a worse starting position than Uber ever occupied — 79% of the public already unwilling to get in — while the livelihoods of roughly 150,000 licensed London drivers, the same workforce that drove nurses through the lockdowns, hang on the outcome. A city with this institutional memory does not need persuading to be careful. It needs a reason not to be, and so far the reason on offer is a press release.

The 2027 question nobody answers at the launch event

When the Automated Vehicles Act's regime matures in 2027, a new legal creature takes the wheel: the authorised self-driving entity, a company that stands in for the driver in law. Ask the launch event what that means for you, a passenger, on the night something goes wrong, and watch the answer dissolve into process. Who attends the scene? Whose insurer leads? How long does a claim take when the "driver" is a corporation in another jurisdiction and the evidence is telemetry the operator controls — the same category of evidence, you will recall, that Cruise was criminally fined for editing? A human driver's accountability is immediate and ambient: it is in the car with you. A corporate entity's accountability arrives by correspondence, in due course, subject to terms. Perhaps the courts will compress that distance in time. They have not yet, anywhere on Earth.

The honest case against us — written by us

An argument you can trust is one that can afford to arm its opponent, so here is the strongest case against human drivers, made without flinching. Humans tire. Humans have moods, blind spots and bad days. Human labour is the most expensive component in your fare, and software does not need rest breaks, pensions or sympathy. Some fraction of human drivers, somewhere, tonight, will drive worse than a median robotaxi in its own geofence. All of this is true, and any operator who pretends otherwise is selling you something.

Here is why it does not settle the question. Every one of those human failings exists inside a system of consequences — licensing, vetting, revocation, livelihood, the courts — that has spent a century learning how to find and remove the worst of us. The robotaxi's failings exist inside a system of patches, marked by the vendor, disclosed at the vendor's discretion, audited — so far — mainly when prosecutors force the issue. We are not claiming humans are flawless. We are claiming that human accountability is load-bearing: it is the structure that makes the flaws survivable. Remove it, and you have not removed error. You have removed the only thing that ever answered for error.

What no robotaxi will ever publish — and we just did

The most powerful content on the internet is the content that is not on it yet. So here is our own record — generated by humans being accountable to humans, and printed where our passengers can hold us to it:

Make it travel — one-click social kit