Driverless taxis arrived in London — minus the part where they’re driverless

The most revealing thing about London’s new driverless taxis is not that they work. They demonstrably do: on a Tuesday morning in early September, a Ford Mustang Mach-E equipped with Wayve’s AI system navigated the city’s organised chaos — its roundabouts, its cyclists threading between buses, its centuries-old streets that no grid planner would ever have designed — and delivered a passenger where they asked to go. Over 140,000 Londoners had already signed up through the Uber app before the service launched. The enthusiasm was genuine.

The revealing thing is what sits in the front seat of each car: a licensed, Transport for London-approved private hire driver, legally required to remain alert and ready to intervene. The UK regulator has not yet approved fully autonomous operation. What Wayve and Uber launched in early September is, in technical terms, supervised autonomy. In commercial terms, it is London’s first driverless taxi service. Both descriptions are accurate. The tension between them is the debate worth having, and it is a debate the launch coverage has largely sidestepped in favour of the milestone framing.

Wayve, the British AI company behind the technology, represents a specific architectural bet. Unlike earlier autonomous driving companies that relied on exhaustive pre-mapping of every street corner, Wayve’s platform learns from experience, the way a human driver does. The company raised $1.2 billion in a Series D round in February 2026, with Microsoft, NVIDIA, Uber, Mercedes-Benz, Nissan, and Stellantis among the investors, at a post-money valuation of $8.6 billion. Total capital secured reached $1.5 billion including milestone-based commitments from Uber tied to deployment scale. The commercial logic is that an AI system capable of adapting to unfamiliar environments without pre-programming can reach new cities at a fraction of the cost that constrained first-generation AV companies. London is where that thesis meets actual streets and actual passengers.

Whether the thesis holds depends on which test you cite. CBS News dispatched a reporter to ride in one of Wayve’s vehicles shortly after launch. The car, initially impressive in navigating the city’s traffic, became stuck in a narrow passageway of the kind that experienced London drivers navigate by feel and social intuition. A safety supervisor had to take control. Neither Wayve nor Uber has disputed the incident; they describe the supervised phase as intentional caution rather than revealed limitation. That distinction, between staged deployment and genuine capability, sits at the centre of the most substantive argument in autonomous vehicle discourse: when a company says its technology is not yet ready to operate without human oversight, is that epistemic honesty or a signal that the claimed capability remains years from the conditions that define commercial viability?

The safety statistics are the most contested territory, and the numbers shift depending on who presents them. Waymo, the Google-backed autonomous vehicle company currently preparing for its own UK launch, published research claiming its vehicles experience 57 percent fewer police-reported crashes than human drivers, with an incident rate of 2.1 per million miles compared to 4.85 for human-driven vehicles. These figures appear in almost every pro-autonomy argument. What travels with them less often is the methodological context: vehicles equipped with autonomous driving systems are subject to significantly stricter incident reporting requirements than human drivers, and the operating conditions — careful route selection, immediate fallback protocols — are not equivalent to full-population driving. Earlier data compiled through 2021 showed autonomous vehicles recording roughly 9.1 crashes per million miles, more than twice the human rate. In 2026, the US National Highway Traffic Safety Administration opened a probe into Waymo after receiving 22 reports of crashes or traffic-law violations; the agency also documented a Waymo recall of 3,871 vehicles after some entered closed freeway construction zones. The safety debate is not settled. Smart analysts with access to the same data reach genuinely different conclusions.

London’s streets make the scale question more acute than in the Sunbelt cities where AV technology has progressed furthest. Wayve’s own leadership has noted that London contains more than ten times the cyclist and pedestrian density of San Francisco — a city whose infrastructure was built in an era of automobiles. A city that grew organically over two millennia, without a grid, with road widths calibrated for horses and Victorian wagons, contains interaction problems that training data from more predictable urban environments cannot fully encode. The CBS test’s narrow passageway was not an edge case engineered to produce failure. It was a Tuesday in central London.

The labour question is the one most consistently underdiscussed in the launch coverage. Black cab driver numbers in London fell from approximately 25,000 in 2010 to around 16,000 today, a decline that predates autonomous vehicles and has multiple causes, prominent among them the arrival of Uber itself. The GMB union, which represents workers across the transport sector, voted at its 2026 annual congress to urge government to introduce statutory protections for taxi and private hire drivers as autonomous vehicle deployment accelerates. The union estimates that approximately 300,000 private hire and taxi driver positions across the UK could face displacement, though this figure is contested and the timeline of actual impact remains uncertain. Ali Haydor, a GMB congress delegate and private hire driver, posed the most direct question in the entire debate: “Progress for whom?”

Uber’s global head of autonomous mobility offered the answer its shareholders prefer: no near-term employment risk, and a growing market capable of absorbing both autonomous and human-driven services. Transport Secretary Heidi Alexander has welcomed the investment, citing government projections of 38,000 new jobs and a £42 billion industry by 2035. The GMB disputes whether those projections account for the displacement of existing workers or represent net new creation, and whether the timeline aligns with drivers’ career spans rather than venture capital return horizons.

The pro-autonomy case deserves to be made properly, because it is not cynical. Autonomous vehicles could provide reliable, accessible transportation for people with disabilities who find the current system inconsistent and often inadequate. They could reduce the approximately 1.35 million road deaths worldwide each year, the large majority caused by human error, if their safety record at scale proves better than the contested current statistics suggest. They could deliver a significant environmental gain through electrification at a moment when the existing taxi fleet cannot decarbonise on a comparable timeline. The government’s 2035 jobs projection, even if optimistic in its framing of net versus gross employment, reflects economic activity that does not currently exist. These are genuine benefits, not marketing language.

The difficulty is structural. Wayve launched with 15 cars in a city of nine million. The gap between those numbers and the fleet scale at which autonomous vehicles begin to meaningfully shift urban mobility is measured in years and regulatory decisions, not months and press releases. Waymo, which has mapped London in preparation for its own planned service, is expected to announce its UK launch timeline before the end of 2026. The competitive pressure between two well-capitalised operators will accelerate deployment pace in ways that may not synchronise with the regulatory process that would determine protections for workers in the transition. That asymmetry — technology moving faster than the governance designed to manage it — is the pattern that has defined every wave of platform-economy disruption over the past fifteen years, from Uber’s own arrival in London to the gig-economy legal battles that followed.

Fifteen cars prove that Wayve’s technology can operate on London’s streets with a human nearby. Fifteen thousand would begin to prove something larger. The journey from one to the other will not happen in political silence, and it will not happen without absorbing costs that are, at this moment, being carried by no one named in any press release. The 140,000 Londoners who opted in were asked. The 300,000 whose livelihoods sit in the adjacent space were not.

What the record confirms — and where the argument is unresolved

What the record confirms: Wayve and Uber launched supervised autonomous ride-hailing in London in early September 2026, the first such service in the United Kingdom. The initial fleet is 15 Ford Mustang Mach-E vehicles. Each operates with a TfL-licensed private hire driver on board under current regulations. The service is available through the Uber app at standard fare rates across UberX, Uber Electric, and Uber Comfort. Wayve raised $1.2 billion in a Series D round in February 2026 with participation from Microsoft, NVIDIA, Uber, Mercedes-Benz, Nissan, and Stellantis. Black cab driver numbers in London have declined from approximately 25,000 in 2010 to 16,000 today.

What remains genuinely unresolved: whether autonomous vehicles are demonstrably safer than human drivers under equivalent operating conditions at scale; how many jobs the technology will displace and over what timeline; whether the UK government’s projection of 38,000 new jobs by 2035 represents net employment or displaced-and-replaced employment; and whether Wayve’s embodied AI architecture outperforms HD-map-based systems in the conditions that define commercial viability rather than supervised proof-of-concept.

Tags: autonomous vehicles, Uber, Waymo, Wayve, driverless taxis