On 3 September 2026, Tesla published its North American Cybercab Rider Guide and compliance and safety documentation. The rider guide explains that users can request driverless Cybercab journeys through the Robotaxi app. Passengers have no conventional driving controls with which to take over, although they can ask the vehicle to pull over using the touchscreen, the mobile app or an overhead stop button. On the same day, the Associated Press reported that dozens of Cybercabs without steering wheels or brake pedals had entered the streets of Austin, Texas, for an invitation-only launch. It remained unclear when the service would become generally available, and the published material did not provide complete operational data from which the system’s real-world safety performance could be independently assessed.
The resulting question is not whether Cybercab is conscious, nor which person an algorithm should choose in a moment of unavoidable harm. It is more specific: when an AI driving system still under development enters a shared public road, who becomes part of that deployment, and on what basis does the system gain legitimate authority to affect them? Invited passengers can decide whether to enter the vehicle. Pedestrians, cyclists, people in other vehicles and emergency workers cannot individually decide whether they will encounter it. Consent to ride governs the relationship inside the vehicle; it does not automatically become the consent of everyone using the road.
Four matters need to be distinguished. Technical operation is a functional fact. Compliance with specified rules or permission to operate is a legal and administrative status. An invited passenger’s willingness to ride is an individual choice. Public legitimacy concerns whether shared institutions can reasonably require all affected people to bear the risks and uncertainties created by the deployment. These four conditions may coincide, but they may also come apart. A completed journey does not by itself establish long-term safety, regulatory permission does not replace continuing oversight, and public unease does not by itself prove that the technology is unsafe.
In his peer-reviewed paper “An Ethical Framework for Evaluating Experimental Technology”, philosopher of technology Ibo van de Poel argues that the social introduction of a new technology can be understood as a social experiment because some effects become visible only through use. He assesses the acceptability of such experimentation through principles including non-maleficence, beneficence, respect for autonomy and justice. This does not make road deployment legally identical to biomedical research, nor does it require every person who might meet a vehicle to sign a consent form. The framework identifies a more basic condition: when consequences can be discovered only in a real environment, developers cannot count the benefits as the achievement of innovation while treating uncertainty as a cost naturally borne by the surrounding public.
Jack Stilgoe and Miloš Mladenović make a related argument in their peer-reviewed article “The Politics of Autonomous Vehicles”. Autonomous driving cannot be understood solely through the algorithm inside the vehicle. It is formed through relations among vehicles, road rules, infrastructure, local institutions and other road users. They particularly note that public-road trials can exclude non-users, bystanders and citizens from the governance of the technology. This bears directly on Cybercab. An invitation process determines who may sit inside the vehicle, but it does not answer who may know the operating boundaries, challenge the distribution of risk, or require rules to be suspended or corrected after an abnormal event.
My judgement is that entering a public road is not the same as earning public legitimacy. Legitimacy does not require zero risk: human driving, road works and public transport cannot eliminate risk either. It requires that new capability and new risk are not defined solely by the deploying organisation, but placed within a public structure that permits scrutiny, accountability, suspension and correction. At a minimum, that structure needs disclosure of vehicle and software versions, actual operating areas, consistent reporting standards for crashes and disengagements, routes for independent review, clear responsibility and compensation arrangements, and a regulator with the practical ability to narrow or suspend operations when new evidence warrants it. Different jurisdictions may implement these functions differently, but they cannot remain only corporate assurances.
In the terms of Sustenesis Theory, the first task is to identify Difference: the distinguishable positions occupied within the same road deployment. The developer controls the system and its data; passengers have a limited choice; other road users bear exposure to interaction; and public institutions carry responsibilities for rules and remedies. Constraint does not mean an obstacle in the ordinary sense. It refers to the conditions that make particular relations possible or impossible, including road rules, geofences, data disclosure, remote support, incident investigation, insurance, compensation and authority to stop operations. Sustained Coherence is not the temporary absence of a failure. It is the continued, testable consistency among technical behaviour, public rules, feedback records and corrective capacity.
On this account, public legitimacy cannot be granted by one launch event, nor permanently completed by one vote or one permit. It is formed and maintained through continuing feedback: which limitations the system reveals, how institutions respond, whether affected people can raise objections, whether incidents change operating conditions, and whether the developer must accept external correction. Without these relations, a vehicle may be technically usable and administratively permitted while still lacking a sufficiently coherent public arrangement.
This also explains why “Does the public trust autonomous vehicles?” is not the best first question. Trust is often framed as an attitude that the public must supply, as though the technology were complete and only social acceptance were missing. The more appropriate sequence begins by asking whether institutions provide conditions worthy of trust. If operating scope, abnormal-event data and correction mechanisms are not visible, demanding trust shifts an institutional responsibility onto individual psychology. Transparent records, independent verification, clear remedies and reversible deployment do not remove disagreement, but they give both trust and distrust grounds that can be examined.
The available evidence is insufficient to judge Cybercab’s actual safety performance, and it does not establish that the Austin operation is unlawful or lacks every form of oversight. The judgement here is limited to a conceptual boundary: product documentation, invited rides and vehicles on public roads establish that deployment has occurred, but they do not by themselves establish public legitimacy. The next relevant evidence is not a general promotional assurance. It is the specific conditions of permission, consistent incident data, external review, public complaint and remedy channels, and whether those forms of feedback can materially change the boundary within which the vehicles continue to operate.
References
Tesla, Cybercab Rider Guide, 3 September 2026, https://www.tesla.com/robotaxi/riderguides/cybercab/en_us/Cybercab-Rider-Guide.pdf
Associated Press, Musk launches steering-wheel-free taxi service in bet riders lose ‘no control’ fears and hop inside, 3 September 2026, https://apnews.com/article/8791add840f4debbc41bcaa9d1e64561
Ibo van de Poel, An Ethical Framework for Evaluating Experimental Technology, Science and Engineering Ethics, 2016, https://pubmed.ncbi.nlm.nih.gov/26573302/
Jack Stilgoe and Miloš Mladenović, The Politics of Autonomous Vehicles, Humanities and Social Sciences Communications, 2022, https://www.nature.com/articles/s41599-022-01463-3
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.