How to Test a Facial Recognition Deployment Against the Proportionality Standard
Face matching is deployed by public authorities faster than the law about it settles, and the question a lawyer is asked is rarely whether it is lawful in the abstract. It is whether this facial recognition deployment, with this legal basis and these safeguards, would survive the standard the Supreme Court applies to an intrusion on privacy.
Step 1: Identify the Right and the Actor
The nine-judge Bench of the Supreme Court held in 2017, in the privacy judgment reported at (2017) 10 SCC 1, that privacy is protected as an intrinsic part of the right to life and personal liberty under Article 21 of the Constitution, and that informational privacy is part of it.
That matters at the outset for two reasons. The standard set out below applies to State action, so the first question is whether the deploying body is the State or is acting for it, including where a private vendor supplies and operates the system. And the right engaged is informational privacy, which is why a deployment that only matches faces against a watch list, retaining nothing, still has to be justified rather than assumed.
Step 2: Ask Whether There Is a Law, Not a Policy
The first requirement is legality: the intrusion must have the backing of law. An administrative instruction, a standing order, a tender document, a memorandum of understanding with a vendor or a budget line are not law for this purpose.
This is where most deployments are weakest and it is the cheapest point to test, because it is answered from documents rather than from argument. Where the authority can point only to an internal approval, the remaining prongs do not need to be reached.
Step 3: State the Aim Precisely Enough to Be Tested
The aim must be a legitimate one, and it has to be stated at a level of particularity that allows the rest of the analysis to proceed. Public order, crime prevention and public safety are legitimate aims in the abstract, but an aim pleaded at that level cannot be tested for rational connection or necessity.
The useful formulation names the harm, the population and the setting: identifying persons already charged in specified offences, at a specified category of location, for a specified period. An aim that cannot be stated that way is usually a purpose looking for a justification.
Step 4: Test the Rational Connection
Suitability asks whether the measure is rationally connected to the stated aim. For face matching this is a technical question before it is a legal one, and it turns on the material the authority holds rather than on the vendor's claims.
The relevant material is the accuracy of the system on the population and conditions in which it is deployed, the rate at which it returns a wrong match, and what happens operationally when it does. A system that produces matches that are then acted on without independent verification is connected to the aim only if the matches are reliable, so the verification step is part of the connection and not a safeguard bolted on later.
Step 5: Test Necessity, Which Means the Alternatives
Necessity asks whether a less restrictive but equally effective alternative was available. This prong is lost more often by omission than by argument, because it requires the authority to show that alternatives were considered.
The alternatives worth putting are specific: targeted surveillance of identified persons rather than scanning everyone present; retrospective matching of recorded footage against a narrow list rather than live matching of a crowd; and conventional identification where the numbers make it practicable. A record that no alternative was examined is itself the answer to this prong.
Step 6: Weigh the Effect, Not Only the Benefit
Balancing asks whether the measure has a disproportionate impact on the right holder. The benefit side is usually documented and the burden side usually is not, so the work here is to articulate the burden.
For live face matching in a public space the burden falls on everyone present rather than on a suspect, it is incurred without notice and without any means of declining, and in settings where people assemble to express a view it attaches a record of attendance to the exercise of a freedom. Those features belong in the assessment whatever conclusion is reached.
Step 7: Look for Procedural Safeguards Against Abuse
The standard requires procedural guarantees against abuse, and this is the prong that can actually be designed. The questions are concrete: what is retained, for how long and on what authority; who may query the system and against which database; whether queries are logged in a form that cannot be edited; who audits the logs and to whom they report; and what a person wrongly matched can do about it.
A deployment that answers those questions in writing is in a materially different position from one that answers none, even where the earlier prongs are contestable.
Step 8: Record the Assessment, and Note What Is Pending
The assessment should be written at the time, prong by prong, with the documents relied on identified. It is evidence of the authority's own reasoning, and for an advisor it is the document that shows which prong the matter will turn on.
It is also worth recording that the law here is moving. The Supreme Court has agreed to hear a challenge to police use of facial recognition and biometric surveillance at protest sites, so an assessment written now should be dated and revisited rather than treated as settled.
Common Pitfalls to Avoid
Treating an internal approval as legal authority: Legality means the backing of law, which an administrative instruction or a procurement document does not supply.
Pleading the aim at the level of public safety: An aim stated that broadly cannot be tested for rational connection or necessity.
Accepting vendor accuracy figures as the connection: What matters is performance in the population and conditions of the actual deployment, and what is done when a match is wrong.
Skipping the alternatives: Necessity requires a record that less restrictive options were considered and why they were rejected.
Counting only the benefit: Balancing requires the burden on everyone scanned, not merely the gain to the authority.
Leaving safeguards to practice: Retention, access control, tamper-evident logging, audit and redress are the prong most capable of being designed, and the easiest to show was not.
Key Statutory Provisions
Article 21 of the Constitution of India: Protection of life and personal liberty. The 2017 privacy judgment of a nine-judge Bench, reported at (2017) 10 SCC 1, holds privacy to be protected as an intrinsic part of the right to life and personal liberty, and informational privacy to be part of that protection.
The proportionality standard applied to an intrusion on privacy: The Court has expressed it in more than one formulation, and the elements common to them are legality, meaning the backing of law; a legitimate aim; suitability or rational connection between the measure and the aim; necessity, meaning the absence of a less restrictive but equally effective alternative; balancing, meaning no disproportionate impact on the right holder; and procedural guarantees against abuse.
Sources and References
The prongs of the proportionality standard, and the two formulations distinguished
Independent statement of legality, legitimacy and the four sub-components
The 2017 judgment, its Bench strength and its reported citation
The Supreme Court agreeing to hear a challenge to police use of face matching at protest sites
Constitution of India, Article 21, and the privacy judgment of 2017 reported at (2017) 10 SCC 1
Disclaimer: This article is for informational purposes only and does not constitute legal advice. Readers should consult a qualified legal professional for advice specific to their circumstances.
Comments