CHAPTER 6

How Privacy Is Protected: VRAIO

VRAIO (Verifiable Record of AI Output) changes the way we think about protecting privacy: govern the output of information, rather than restricting measurement.

What do we really fear?

Security technologies such as encryption have advanced greatly against outside attacks. But misuse of information by insiders with legitimate authority remains a fundamental problem.

Much of the concern about privacy is not about being recorded as such, but the fear that recorded information will be used, without our knowledge, for improper purposes. If so, the critical point to protect is the exit through which information leaves the system.

How VRAIO works

Set the rules: who may retrieve which information, for what purpose, about which subject and time range, and to whom it may be passed. Society sets the rules.
Declare: anyone who wants to retrieve information declares the purpose, the content, its compliance with the rules and the grounds, in a common format (GLO: Guide to the Expression of Legitimacy of Output).
Record: an independent Recorder checks the declaration against the rules and records the declaration and the disclosure decision in tamper-evident form. The Recorder never needs to see the footage or hold the decryption keys.
Audit: after disclosure, spot audits compare declarations with the information actually retrieved, detect misconduct and identify those responsible.
Sanction: misconduct receives effective sanctions. The system is designed so that the cost of being detected and sanctioned outweighs any benefit of misuse.

What this looks like in FMPS

Operation and key management: carried out by the municipality. Equipment manufacturers and telecom operators cannot freely view footage.
Requesters: the police (investigations) and municipal departments (for example, searches for missing persons).
Retrieval by police: one possible approach is to require a court warrant normally, and to obtain the warrant afterwards in emergencies.
Emergencies: declaration, recording and audit follow the same procedure as in normal times.
Granularity: every viewed image is recorded individually.

Figure: FMPS + VRAIO architecture. Footage is stored encrypted at each streetlight and can be retrieved only through a disclosure procedure recorded by the Recorder.

By mechanism, not goodwill

Surveillance cameras have so far relied on the goodwill of operators and compliance with guidelines. VRAIO builds, as a mechanism, a state in which misconduct is detectable and sanctioned. For honest operators, it is also a way to demonstrate their legitimacy.

The prototype of this idea has already been operated as a real system. In the e-JIKEI Net Camera field trials in Kiryu City, footage could not be viewed without permission from the recording server, and every viewing was made public (Chapter 11; Demo 3: Stalking by a city-hall employee (PDF)).

Society decides the rules

VRAIO does not decide the specific rules: who may retrieve what, whether a warrant is required, whether the person viewed is notified, how usage statistics are published. These are for society to decide through laws and ordinances. VRAIO is a mechanism that makes it possible to verify afterwards that the agreed rules were actually followed.

Not protection by lowering image quality

Some feel reassured by low-quality cameras that cannot capture faces. In FMPS, however, image quality is set by what tracking and identification require; at key locations, faces and number plates are recorded clearly. Privacy is protected not by reducing measurement capability but by governing outputs. This is the core of VRAIO.

← Previous: 5. FMPS in Numbers | Next: 7. If a Government Tried to Misuse It →

FMPS + VRAIO: Public-safety infrastructure built on streetlights, and the governance of AI outputs | Yusaku Fujii (Gunma University) and NPO e-JIKEI Network Promotion Institute
This site presents a research-stage concept; all figures are estimates.