Memorandum submitted by Caspar Bowden
Privacy Enhancing Technologies (PETs) span disparate
fields of computer science research[260]
(eg the cryptography of "private credentials", untraceable
and unlinkable data transport mechanisms, biometric encryption,
human-computer interaction and usability, policy control languages,
database/statistical privacy) which in combination can create
complete "privacy systems". PETs should not be considered
to be a "toolbox" which can rectify specific privacy
problems in isolation.
Industry and academia are able and willing to
develop effective PETs and privacy systems, but there is a chronic
lack of awareness and interest from both data controllers and
most regulators. Since data protection compliance obligations
fall on data controllers, in the absence of clear incentives (regulatory
or economic) to deploy PETs, it is unreasonable to expect them
to become widespread through market forces or regulator exhortation
alone. Sanctions sufficient to deter organisations from treating
breaches of information privacy as an acceptable business contingency,
and greater public awareness of privacy risk are probably necessary
to generate market demand.
Design issues in identity systems and privacy
systems are really complementary and inseparableand they
require "threat modelling" of risks both to the individual
and organisationswhich are not necessarily similar or symmetrical.
However the DP principle that processing should not be "excessive"
in relation to specified and legitimate purposes is crucial for
reconciling conflicting interests, but the policy of the ICO to
date has generally eschewed major interventions to halt disproportionate
processing,[261]
in contrast to some other European DPAs. The policy climate for
discussion of enterprise information security in the US has been
profoundly affected by the broad scale of security breaches that
have come to light, since the passage of security breach disclosure
laws, however the ICO remains ambivalent whether individuals have
a right to be informed.
Privacy is often fatally compromised in designs
for large scale identity systems through preconceptions that all
transactions need ultimately to be traceable or (re-) identifiable,
extreme hypothetical cases used to justify overbroad processing
of personal information, and reliance on procedural rather than
technical safeguards. Sophisticated PETs can provide much more
robust bulwarks against function-creep than policy controls alone,
but it must be understood that the purpose of these technologies
is expressly to minimize the disclosure of personal information
to the absolute minimum required.
Advanced PETs for identity management have some
very curious properties, which have been painstakingly designed
by a small number of world-class cryptographers over the past
twenty years, to try and achieve better outcomes for both security
and privacy. However the subtlety of the problems these techniques
are designed to solve can seem very abstruse to non-specialists,
and the techniques themselves can accomplish things which might
seem logically impossible, or contrary to the intuition of the
layperson.
For example, it is possible to authenticate
a transaction in such a way that if a cryptographic token (which
proves entitlement to some service) is used only once (ie honestly
and as intended), it can be mathematically guaranteed that the
individual cannot be identified. However if an attempt is made
to forge or copy a token, or use it more than once, then only
in that contingency does it become possible to trace and identify
the person to whom it was issued.
The above idea was essentially conceived as
a way of implementing online payment services with the privacy
properties of cash. A more recent and much more widely applicable
technique is the ability to revoke the validity of long-lived
untraceable tokens with service providers other than those that
issued the tokens, without the necessity to identify the owner
of the token.
There is now a family of such innovative cryptographic
techniques which together constitute a very powerful new paradigm
for combating fraud and abuse, whilst strongly preserving the
privacy of honest users. It means that privacy is compromised
if and only if dishonesty is detected, and thus potentially forms
the basis of a new kind of social contract with citizens. The
potential applications include:
road-pricing and congestion
charging;
welfare benefits, healthcare,
and social services entitlement;
private sector use of data from
the National Identity Register; and
use of a national identity card
in over-the-counter transactions.
The fundamental legal and policy issue this
raises is that if one takes the Human Rights Act (and Art.8.1
of ECHR) seriously, the state has a duty to limit intrusions into
privacy to that which is necessary, not in a general sense, but
case-by-case according to the circumstances of the individual.
Therefore the use of these advanced PET techniques is mandated
by the HRA (subject to reasonable feasibility), because it infringes
privacy only to an extent that is individually proportionate.
This is in stark contrast to schemes (such as the Oyster card)
which rely on blanket collection of identifiable transaction data,
and thus are highly vulnerable to "function creep".
As things stand, in systems which collect all
transactional data identifiably (on the basis that it cannot be
predicted in advance which transactions may need to be retrospectively
traced for fraud investigation), a "side-effect" is
that a database of all transactions is retained (for some period),
but because the database exists, it is a temptation to use it
for general surveillance or other purposes unrelated to its primary
function.
However, to see the connection between human
rights and these advanced technologies, one has to appreciate
both the counter-intuitive possibility of such "conditional
identifiability", and the implications of existing ECHR jurisprudence.
So far, no parliamentary inquiry in any ECHR jurisdiction has
spanned this legal and technical gulf. Policy makers are simply
unaware these technical possibilities exist. Thus the legality
of blanket retention of identifiable transaction data is never
fundamentally questioned, because there seems to be no logical
alternative to providing a realistic capability for audit and
fraud control.
This is the main point I would wish the Committee
to consider in this inquiry. However I would also make the following
recommendations for specific reform of the Data Protection Act
1998.
The right of the data subject
to access their personal data should in general be exercisable
online and without charge. I have more detailed proposals for
the necessary technical and policy reforms.
The definition of personal data
in S1, which presently might exclude data which is only "indirectly"
identifiable from being personal, should be altered to implement
fully Recital 26 of the Directive ("or by any other person").
There should be a presumption
that the consent of the data subject is required for processing
personal data, with the onus on the controller to specify why
derogation from obtaining consent is justified. Essentially the
emergence of user-centric identity management technologies makes
this feasible. In previous pre-Internet decades of data protection
policy it would not have been feasible.
Disclosure: although I am submitting
this memo as a private individual, and not to represent the views
of my employer, I feel it is proper to disclose that between the
time this note was initially drafted and later finished, Microsoft
has (partially in consequence of my recommendations) acquired
the intellectual property of Credentica Inc, and hired Dr Stefan
Brands, one of the leading cryptographers in the field of advanced
PETs.
8 April 2008
260 Digital Privacy: Theory, Technologies, and Practices:
Alessandro Acquisti, Stefanos Gritzalis, Costos Lambrinoudakis,
Sabrina di Vimercati eds, Auerbach 2007. Back
261
The affair of fingerprinting at Heathrow Terminal 5 is a rare
exception. Back
|