An elderly woman who has worked the fields all her life goes to the ration shop for her month’s grain. The shop authenticates entitlement by fingerprint, and decades of manual labour have worn the ridges of her fingers smooth, so the scanner does not recognise them. She tries again, and it still does not. She goes home without her rations.
The usual way to describe what went wrong is in the language of consent and data protection: whether she agreed to biometric authentication, whether she was told how her data would be used and what alternatives she had, whether she could have opted out. These questions organise the Aadhaar Act’s own provisions, the Supreme Court’s judgments, India’s Data Protection Act of 2023, and much of the activism and scholarship that has grown up around the system’s failures. I think they are the wrong questions for her case, and the reason they are wrong tells us something about what algorithmic welfare systems are.
Not a failure to consent
She is not failing to give informed consent; she is failing to be recognised by the system as someone who could give or withhold consent to anything at all. Her relationship to Aadhaar is not that of a rights-bearer, already standing, who is then asked to authorise the processing of her data. Aadhaar is the condition under which she becomes visible to the welfare state as a rights-bearer in the first place, and she cannot refuse the system without refusing the standing from which a refusal would make sense.
The point needs stating carefully, because it is easily heard as something absurd. She is a person, a citizen and a legal subject whatever any database says, and her human rights are not the creation of an identity authority. What the system produces is narrower and, for her purposes, decisive: her operative standing at a particular counter, the standing in virtue of which she counts there as someone to whom grain is owed and is handed over. When the fingerprint fails, that standing fails with it.
I call this condition constitutive dependence. A person stands in it when an institution produces her operative recognition as a bearer of entitlements rather than merely administering a recognition that exists independently of it; when she has no standpoint from which to refuse or correct it without giving up the standing she would be protecting; and when the institution can revise her recognition through its own operations, without appeal to an independent ground.
Why consent cannot do the work
The liberal tradition asks four things of valid consent. The person must be competent, informed and acting voluntarily, and she must have standing over the matter, the kind of standing that makes her authorisation count. Debates about consent going wrong almost always concern the first three: the patient without capacity, the user who did not read the policy, the worker who cannot refuse surveillance. The fourth is taken for granted.
Her case fails on the fourth. For a documented, urban citizen, refusing to enrol is at least imaginable: she can weigh the costs of living outside the system and decline them. For a rural labourer without a birth certificate or a stable address, the standpoint of refusal is illusory, and not only costly. In refusing Aadhaar she would refuse her existence as someone the welfare state can see, not an identification method, and to count that as a high price for an option is a category mistake about what kind of option is on offer.
This distinguishes two ways consent can fail. Sometimes it fails practically: its conditions could be met but are not, as when a privacy policy is too long to read, and better design is the fix. Sometimes it fails conceptually, and no redesign, however ambitious, could restore it. Medical ethics already recognises the second kind. When a patient lacks the capacity to consent, we do not print a clearer leaflet; we turn to surrogate decision and best interests. Constitutive dependence calls for a comparable shift, though not because the people affected lack capacity: the standing that consent presupposes is what the system produces.
Three faces of the same structure
The woman at the ration shop is one face of it. A second is the migrant worker, whose entitlement to subsidised grain has historically been tied to a particular shop in a particular district, so that the movement on which her livelihood depends puts her legibility, and with it her food, at risk. Schemes for national portability help many and miss others, especially those whose documents are incomplete at one end or the other. The system does not record a geography of entitlement that exists independently of it; it makes one.
A third is the child. A child enrolled in Aadhaar before she can agree to anything has her standing as a recipient of school meals and scholarships produced, in part, through the enrolment itself, and no proxy consent by her parents can be authorisation in the full sense. What the child shows openly is true, less visibly, of every adult the system governs.
None of the three cases depends on a contested number. Estimates of how often authentication fails vary across surveys and are disputed by the authority itself, and I take no stand on their magnitude. The argument needs only what is not seriously in dispute: that failure is patterned rather than random, concentrated among people whose hands carry the marks of the labour the welfare system exists to protect, and that for those it excludes the consequences can be grave.
What such a system owes
If consent cannot legitimate a system that produces its subjects’ standing, something else must. My proposal, which I call the Constitutive Obligation Principle, is that the legitimacy of such a system depends on discharging obligations its subjects cannot waive, since waiving would itself be a form of the authorisation that is not available. There are at least three, each answering to a feature of the dependence.
Legibility. Because the institution produces recognition, it cannot treat its criteria as natural facts beyond its responsibility. A fingerprint criterion that systematically fails the hands of manual workers is a choice, and the institution owes its subjects criteria that fit the material reality of their lives.
Non-abandonment. Because the institution can revise recognition through its own operations, it may not withdraw what it produced without meeting a burden of justification. A cancelled card or a failed database link is a withdrawal of recognition, not a clerical event.
Remediation. Because its subjects have no standpoint outside it, they must have a substantive pathway inside it when recognition fails. Sending the woman whose fingerprints failed to an online grievance portal discharges no genuine obligation; it redoubles the failure.
Not an outlier
It would be comforting to treat Aadhaar as a peculiarity of one country’s administration. I think it is an early instance of a form towards which algorithmic governance is converging: automated decisions, criteria enforced without the discretion by which an official can recognise someone manifestly entitled, and entitlements conditioned on records held and cross-checked elsewhere. Wherever those three features come together, some people’s standing will be produced by the system rather than presupposed by it. For them the question is what the system owes them, and whether it is paying, rather than whether they agreed.