Trusted Cells: A Sea Change for Personal Data Services

Abstract

How do you keep a secret about your personal life in an age
where your daughter’s glasses record and share everything she
senses, your wallet records and shares your financial transactions,
and your set-top box records and shares your family’s energy
consumption? Your personal data has become a prime asset for
many companies around the Internet, but can you avoid -- or even
detect -- abusive usage? Today, there is a wide consensus that
individuals should have increased control on how their personal
data is collected, managed and shared. Yet there is no appropriate
technical solution to implement such personal data services:
centralized solutions sacrifice security for innovative applications,
while decentralized solutions sacrifice innovative applications for
security. In this paper, we argue that the advent of secure
hardware in all personal IT devices, at the edges of the Internet,
could trigger a sea change. We propose the vision of trusted cells:
personal data servers running on secure smart phones, set-top
boxes, secure portable tokens or smart cards to form a global,
decentralized data platform that provides security yet enables
innovative applications. We motivate our approach, describe the
trusted cells architecture and define a range of challenges for
future research.

Full text

Trusted Cells: A Sea Change for Personal Data Services
Nicolas Anciaux1, 2, Philippe Bonnet3, Luc Bouganim1, 2,
Benjamin Nguyen1, 2, Iulian Sandu Popa1, 2, Philippe Pucheral1, 2
1 INRIA Paris-Rocquencourt
Le Chesnay, France
<Fname.Lname>@inria.fr
2 PRISM Laboratory
Univ. of Versailles, France
<Fname.Lname>@prism.uvsq.fr
3 IT University of Copenhagen
Copenhagen, Denmark
phbo@itu.dk
ABSTRACT
How do you keep a secret about your personal life in an age
where your daughter’s glasses record and share everything she
senses, your wallet records and shares your financial transactions,
and your set-top box records and shares your family’s energy
consumption? Your personal data has become a prime asset for
many companies around the Internet, but can you avoid -- or even
detect -- abusive usage? Today, there is a wide consensus that
individuals should have increased control on how their personal
data is collected, managed and shared. Yet there is no appropriate
technical solution to implement such personal data services:
centralized solutions sacrifice security for innovative applications,
while decentralized solutions sacrifice innovative applications for
security. In this paper, we argue that the advent of secure
hardware in all personal IT devices, at the edges of the Internet,
could trigger a sea change. We propose the vision of trusted cells:
personal data servers running on secure smart phones, set-top
boxes, secure portable tokens or smart cards to form a global,
decentralized data platform that provides security yet enables
innovative applications. We motivate our approach, describe the
trusted cells architecture and define a range of challenges for
future research.
1. INTRODUCTION
With the convergence of mobile communications, sensors and
online social networks technologies, we are witnessing an
exponential increase in the creation and consumption of personal
data. Paper-based interactions (e.g., banking, health), analog
processes (e.g., photography, resource metering) or mechanical
interactions (e.g., as simple as opening a door) are now sources of
digital data linked to one or several individuals. They represent an
unprecedented potential for applications and business.
Until now, the enthusiasm for new opportunities has thwarted
privacy concerns. Nevertheless, the risk of a backlash is growing
as new devices and new services bring us closer to the dystopias
described in the science fiction literature. This risk is well
documented and the nature of the solution is consensual: it is
necessary to increase the control that individuals have over their
personal data [11,9,12]. The World Economic Forum even
formulates the need for a data platform that allows individuals to
manage the collection, usage and sharing of data in different
contexts and for different types and sensitivities of data [13].
Unfortunately, none of the solutions available today can be used
to implement this vision. Centralized solutions, including
emerging cloud-based personal data vaults management
platforms1, trade security and protection for innovative services.
At best, such approaches formulate sound privacy policies, but
none of them propose mechanisms to automatically enforce these
policies [1]. Even TrustedDB [3], which proposes tamper-resistant
hardware to secure outsourced centralized databases, does not
solve the two intrinsic problems of centralized approaches. First,
users get exposed to sudden changes in privacy policies. Second,
users are exposed to sophisticated attacks, whose cost-benefit is
high on a centralized database.
Decentralized solutions are promising because they do not exhibit
these intrinsic limitations. However, existing decentralized
solutions sacrifice functionality or usability for security. Many
examples are discussed in [8]. Other examples include the PDS
vision [2] or the FreedomBox [4]. In PDS, a personal data server
is embedded in a tamper-resistant portable token to hold the
personal data of a user, but the sharing of data is cumbersome
(since the tokens are mostly disconnected) and the range of
personal services is limited (since the tokens have extreme
resource constraints). FreedomBox aims at providing a software
platform that interconnects groups of individuals that trust each
other, thus drastically limiting the range of services it can support.
We argue that the advent of secure hardware embedded in all
forms of personal devices, at the edges of the Internet, will trigger
a sea change. Recently, AMD announced that it will incorporate a
secure Trust Zone-based2 ARM processor on its chips to be
included into smart phones, set-top boxes and laptops. Such
secure tamper-resistant microcontrollers provide tangible security
guarantees in the context of well-known environments3. We can
now imagine that whenever you take a picture, your smart phone
securely contacts the personal services of all individuals in the
frame of the picture, and automatically blurs the face of those who
request it. We can also imagine that the GPS tracker in your son’s
car gives him detailed turn-by-turn guidance, but hides those
details to local government, only delivering road-pricing results.
In this paper, we propose the vision of trusted cells, i.e., personal
data servers running on secure devices to form a decentralized data
platform. We illustrate how trusted cells can be used in the context
of an application scenario, describe the trusted cells architecture
and discuss requirements and challenges for future research.
1 These include Personal (http://www.personal.com), My personal vault
(http://www.mypersonalvault.com), or Mydex (http://www.mydex.org).
2 http://www.arm.com/products/processors/technologies/trustzone.php
3 The adoption of a standard API for secure micro-controllers [5] and the
availability of an open source embedded secure operating system based
on it (Open Virtualization) now enable higher level services.
This article is published under a Creative Commons Attribution License
(http://creativecommons.org/licenses/by/3.0/), which permits distribution
and reproduction in any medium as well allowing derivative
works, provided that you attribute the original work to the author(s)
and CIDR 2013.
6th Biennial Conference on Innovative Data Systems Research (CIDR ’13)
January 6-9, 2013, Asilomar, California, USA.

2. MOTIVATION
Alice lives in France with Bob and their two children. Their house
is now one of the 35 million households equipped with a Linky
power meter. The power meter reports once a day to the
distribution company, a certified time series of readings for
verification, billing and network operation [6]. Alice and Bob
have installed an energy butler app on their secure home gateway,
a trusted cell managing all smart appliances in their home and
storing their data. That award-winning app relies on external feeds
from their utility and local weather prediction, as well as a feed of
readings received every second from the Linky4, to control their
heat pump and the charge of their electrical vehicle. This app
minimizes overall load on the distribution network and saves them
30% on their bill. In addition, Alice is engaged in a social game (a
follow-up to simpleEnergy.com) where she competes with some
friends on their energy savings, reducing consumption by 20%.
At the 1Hz granularity provided by the Linky, most electrical
appliances have a distinctive energy signature. It is thus possible
to infer from the power meter data which activities Alice and Bob
are involved in at specific points in time [7]. How do Alice and
Bob configure the home gateway trusted cell to preserve privacy
while preserving the benefit of their applications? They have a
shared account on this trusted cell. Bob, Alice and their children
have agreed that they do not want to fully disclose all their
activities to each other. They rather have access to 15 min
aggregates via a visualization app – at that granularity one cannot
detect specific activities, but it is still possible to infer a daily
routine. At the same time, daily statistics feed their social game,
monthly statistics are delivered to the distribution company and
time series at required granularity are securely exchanged with
other trusted cells in their neighborhood to achieve consumption
peak load shaving.
None of this data leaves the trusted cell application unless it is
accessed via a predefined set of aggregate queries. The trusted cell
guarantees that no malware can tamper with the data. If the trusted
cell gets stolen, an elaborate attack would need to be mounted to
break the secure hardware and get access to their personal data.
This scenario can be easily transposed to different types of
personal data like GPS traces, Internet traces, mobile phone data,
bills, pay slips, photos as well as health, administrative or scholar
records. We classify the data that could be managed with trusted
cells, based on how and who actually produces it:
(1) Data produced by smart sensors installed by companies in the
user’s home (e.g., power-meter, heat sensor) or in the user’s
environment (e.g., user’s car GPS tracking box for a PAYD
application) on which the user has full or shared ownership,
externalizing aggregated data. Users may opt-in for small-
scale sharing (e.g., local traffic optimization) or larger-scale
sharing (e.g., social games or traffic optimization).
(2) Data produced or inferred by external systems (e.g., purchase
receipt obtained by near field communication or medical data
sent by the hospital or labs). Small-scale sharing allows the
user to optimize her buying habits or to compare her medical
treatment with people having the same disease. Larger-scale
4 In France, such a short-range radio link is a requirement from the
regulation authorities. In other countries, the data from a smart meter
might not be directly accessible. In the US for example, the Green
Button initiative allows customers to obtain online the smart meter data
collected by their utility (http://www.greenbuttondata.org/)
sharing brings public health insights (e.g., epidemiological
study cross-analyzing diseases and alimentation).
(3) Data authored by the user herself (e.g., a photo, a mail, a
document) on which she has complete ownership. Small-scale
sharing benefit is obvious here. Larger-scale sharing of partial
data (e.g., photo location only, number of exchanged mails) is
undoubtedly a source of precious information (e.g., most
interesting places on Google maps).
3. TRUSTED CELLS ARCHITECTURE
What personal data services actually run on a trusted cell? How
do these services allow a user to control whom she shares her
secrets with? How do applications access these services? What
kind of guarantees do trusted cells offer about the security of the
data they manage? We obviously do not aim at answering those
questions fully in this paper. Our goal here is to draw the contours
of an architecture based on Trusted Cells interconnected via an
Untrusted Infrastructure.
Trusted Cells: A trusted cell implements a client-side reference
monitor [10] on top of secure hardware. At a minimum, the
hardware must guarantee a clear separation between secure and
non-secure software. We abstract a Trusted Cell as (1) a Trusted
Execution Environment, (2) a tamper-resistant memory where
cryptographic secrets are stored, (3) an optional and potentially
untrusted mass storage and (4) communication facilities.
Physically, a trusted cell can either be a stand-alone hardware
device (e.g., a smart token) or be embedded in an existing device
(e.g., a smartphone based on ARM’s TrustZone architecture).
The very high security provided by trusted cells comes from a
combination of factors: (1) the obligation to physically be in
contact with the device to attack it, (2) the tamper-resistance of
(part of) its processing and storage units making hardware and
side-channel attacks highly difficult, (3) the certification of the
hardware and software platform, or the openness of the code,
making software attacks (e.g., Trojan) also highly difficult, (4) the
capacity to be auto-administered, contrary to high-end multi-user
servers, avoiding insider (i.e., DBA) attacks, and (5) the
impossibility even for the trusted cell owner to directly access the
data stored locally or spy the local computing (she must
authenticate and only gets data according to her privileges).
In terms of functionality, a full-fledged trusted cell should be able
to (1) acquire data and synchronize it with the user’s digital space,
(2) extract metadata, index it and provide query facilities on it, (3)
cryptographically protect data against confidentiality and integrity
attacks, (4) enforce access and usage control rules, (5) make all
access and usage actions accountable, (6) participate to
computations distributed among trusted cells. Basic (e.g., sensor-
based) trusted cells may implement a subset of this.
Untrusted infrastructure: The infrastructure provides the
storage, computing and communication services, which expand
the resources of a single trusted cell and form the glue between
trusted cells. By definition, the infrastructure does not benefit
from the hardware security of the trusted cell and is therefore
considered untrusted. We consider that the infrastructure is
implemented by a Cloud-based service provider5.
In terms of functionality, the untrusted infrastructure is assumed
to: (1) ensure a highly available and resilient store for all data
outsourced by trusted cells, (2) provide communication facilities
5 A P2P infrastructure among trusted cells could be envisioned but would
raise many technical issues of limited interest for this article.

Figure 1: Alice (A) and Bob (B) are equipped with fixed and
portable trusted cells, acquiring data from several data sources,
synchronizing with their encrypted personal digital space on the
cloud. Charlie (C) is travelling around the world and can securely
access all his data from any (unsecure) terminal thanks to his
portable trusted cell. All users equipped with trusted cells can
securely share their encrypted data through the cloud.
among cells and (3) participate to distributed computations (e.g.,
store intermediate results), provided this participation can be
guaranteed harmless by security checks implemented at the
trusted cells side.
Figure 1 illustrates how trusted cells and the untrusted
infrastructure can collaborate to implement scenarios meeting the
privacy requirements stated above.
Threat model: In our context, the primary adversary is the
infrastructure. The infrastructure may deviate from the protocols it
is expected to implement with the objective to breach the
confidentiality of the outsourced data. Integrity attacks (e.g., on
data related to access control) must also be deterred since they
may lead to subsequent confidentiality leaks. The infrastructure is
assumed trying to cheat only if it cannot be convicted as an
adversary by any trusted cell. Indeed, revealing a data leak (or a
denial of service) in a public place would cause irreversible
political/financial/legal damage to the service provider. Such
adversaries are usually called malicious adversary having weakly
malicious intents [14]. Trusted cells are themselves presumably
trusted. However, even secure hardware can be breached, though
at very high cost, so that one cannot exclude with certainty that a
very small number of trusted cells be compromised. Hence, the
trusted cells’ cryptographic secrets must be managed in such a
way that a successful attack on a (small set of) trusted cell cannot
degenerate in breaking class attack. This is of utmost importance
considering also that an individual succeeding in breaking her
trusted cell could have effective malicious intents.
4. REQUIREMENTS AND CHALLENGES
We identify five major requirements for the user to actually
control how the data entering her personal digital space is
collected, protected, shared and finally used.
Controlled collection of sensed data: The targeted user(s) should
be the unique recipient(s) of raw sensed data and would accept
externalizing only aggregates by opting in/out for selected
applications/services.
At home, the power meter continuously pushes raw measurements
to Alice’s and Bob’s trusted cell gateway, while a certified
aggregated time series is sent to the power supplier company and
aggregates for a social game are pushed to the Cloud every day.
Similarly, the tracking box installed on Alice’s car is a trusted cell
delivering aggregated GPS data to her insurer and raw data to her
trusted cell smartphone that she will synchronize with her
personal space for further use when back home. Hence, adding a
trusted cell to a sensor, allows defining e.g., the frequency and or
precision of the data that should be externalized, thus leading to a
trusted source both for the user (in terms of privacy preservation)
and the provider (in terms of certification of the output data).
Related challenges: Co-design is a primary issue to allow the
definition of affordable sensor-based trusted cells. Low-cost is
indeed a prerequisite to the generalization of trusted sources,
capable of securely filtering and aggregating stream-based spatio-
temporal data with tiny hardware resources. Some trusted sources
being weakly connected to the Internet; asynchrony problems
must also be addressed. Finally, the combination of data streams
from multiple sources, each being separately harmless, may
generate new privacy risks that must be carefully tackled.
Secure private store: All data must be made highly available,
resilient to failure and protected against confidentiality and
integrity attacks. Accessing this data from any terminal, including
those outside the user’s ownership sphere (e.g., internet café),
should leave no trace of the access.
Cryptographic techniques (i.e., encryption, hashing, signatures)
are used to protect trusted cell’s data, keeping cryptographic keys
in their tamper-resistant memory. The data is then stored in the
Cloud and potentially cached in the trusted cell local mass
storage. At a minimum, trusted cells keep locally extended
metadata: access information, indexes, keywords, and
cryptographic keys. Metadata should be sufficient to allow
performing queries before accessing the Cloud to retrieve the data
of interest. Cryptographic keys never leave the trusted cells
tamper-resistant memory. Hence a trusted cell can be used to get
securely data from any (untrusted) terminal it is connected with.
Related challenges: Designing an intuitive HCI for managing this
bunch of heterogeneous personal data (data modeling, data
integration, querying) is a major challenge. Besides, a significant
amount of data and metadata is likely to be embedded in some
trusted cells and may need to be queried efficiently. While it does
not seem a major issue in powerful trusted cells (e.g., a smart
phone), it appears much more challenging when facing low-end
hardware devices like secure tokens (e.g., a microcontroller with
tiny RAM, connected to NAND Flash chips or SD cards, possibly
with energy consumption constraints). Whatever their complexity,
trusted cells should also be designed to support self-tuning, self-
diagnosis and self-healing to minimize the management burden
put on the trusted cell owner.
Secure sharing: The user can decide to keep her data private or
share it with other users or group of users under certain conditions
(e.g., time, location). Under which model the access control
policies are actually defined is an open issue, but not the main
concern of this paper. However, we insist that the user must get a
proof of legitimacy for the credentials exposed by the participants
of a data exchange and must trust the evaluation of the exchange
conditions (if any).
A
A & B services
A
A
Power
provider
Super-
market
Car insurer
B
Sync.
A
Portable
Trusted Cell
Encrypted
personal vault
Secure data
exchange
Photos
MyFiles
Sync.
Fixed Trusted Cell
Sync.
Heat
sensor
Power
meter
F E
D H
G
I
C Internet
café
A & B home
D
E
F
H
I
G
B
A
Employer
Hospital
School
C
C
GPS
Pay slips
Medical data
Scholar folder
The cloud

Practically, sharing data means sharing the associated metadata
(so that the recipient user can get the referenced data in the
Cloud), the cryptographic keys (so that her trusted cell can
decrypt them) and the sticky policy (so that her trusted cell can
enforce the expected access control rules). Hence, thanks to its
security properties, including the protection against illegitimate
actions of the recipient user, the recipient trusted cell can enforce
all the conditions appearing in the access control rules (user’s
credential, contextual conditions).
Related challenges: Again, an intuitive HCI for defining the
access control policies and simple modes of operation must be
devised. The trusted cells themselves may be a source of
simplification (e.g., integration of biometric sensors to
automatically authenticate users, automatic production of certified
credentials safely computed on a trusted cell, definition of default
policies by trusted third parties – e.g., citizen associations – which
could be automatically selected depending on a computed
individual’s profile). Also, secret management is at the heart of
any sharing protocol between trusted cells (i.e., at this level a
secret is a cryptographic key) and must be carefully designed
(e.g., class-breaking attacks must be prevented, master secrets
must be restorable in case of crash/loss of a trusted cell).
Secure usage and accountability: Usage control usually refers to
UCONABC [8]: obligations (actions a subject must take before or
while it holds a right), conditions (environmental or system-
oriented decision factors), and mutability (decisions based on
previous usage)6. Again, defining appropriate usage control
policies for trusted cell applications is an open issue.
Similarly to access control rules, usage control rules can be
implemented as sticky policies so that they are made
cryptographically inseparable from the data to be protected.
Hence usage control rules will be enforced by any trusted cell
downloading data and cannot be bypassed by the recipient user.
Regarding accountability, the recipient trusted cell can maintain
an audit log, encrypt it and push it on the Cloud to the destination
of the originator trusted cell.
Related challenges: Many challenges are common with secure
sharing. However, trusted cells hold the promise of new usages
and new usage controls. For example, trusted cells could be
parameterized so that any personal data produced by a trusted
source linked to an individual A and referencing individual B be
submitted for approbation to B’s trusted cell before being
integrated to A’s digital space.
Shared Commons: Privacy has also a collective dimension in the
sense that preserving one’s privacy should not hinder societal
benefits (e.g., census, epidemiologic releases, global queries). A
trusted cell user is thus expected to participate to global treatments
assuming her data suffers appropriate transformations (e.g,
anonymization, output perturbation) depending on the
trustworthiness of the recipient(s) and the expected usage of the
data/query. When data needs to be transformed before being
delivered, the recipient trusted cell implements the transformation
on its own if possible (e.g., filtering, local data perturbation) or in
collaboration with other trusted cells if the transformation requires
a collective action (e.g., anonymization, global data perturbation).
In the latter case, the computation may be implemented in a pure
Secure Multi-Party fashion or may require the participation of the
untrusted infrastructure (e.g., to store intermediate results).
6 For instance, a photo could be accessed ten times (mutability), in the
course of 2012 (condition), informing the owner of the precise access
date (obligation).
Related challenges: Such large scale computations may lead to
atypical distributed protocols combining security and performance
requirements in an asymmetric context made on one side of a very
large number of highly secure, low power and weakly available
trusted cells and on the other side of a highly powerful, highly
available but untrusted infrastructure. Hence, the trusted cells
architecture can be seen as a massive untrusted interconnection of
trusted co-processors.
5. CONCLUSION
We proposed the trusted cell architecture, a vision reconciling
individual’s privacy with innovative acquisition and sharing of
personal data. This vision is based on the premise of ubiquitous
and open secure hardware. Trusted cells enforce access and usage
control at the edges of the Internet, and thus constitute a sea
change with respect to personal data management. This vision
undoubtedly opens a set of exciting challenges that must be
explored by the database community.
6. ACKNOWLEDGEMENTS
This work has been partially funded by the French ANR KISS
project.
7. REFERENCES
[1] R. Agrawal, J. Kiernan, R. Srikant, Y. Xu: Hippocratic
Databases. VLDB 2002: 143-154
[2] T. Allard et al.: Secure Personal Data Servers: a Vision
Paper. PVLDB 3(1): 25-35 (2010)
[3] S. Bajaj, R. Sion: TrustedDB: a trusted hardware based
database with privacy and data confidentiality. SIGMOD
Conference 2011: 205-216
[4] FreedomBox: http://freedomboxfoundation.org/.
[5] Global Platform Device Technology. Trusted Execution
Environment Internal API Specification. Version 1.0.
December 2011.
[6] S. Katzenbeisser and K. Kursawe, Privacy and Security in
Smart Energy Grids, Dagstuhl Seminar 1151, 2011
[7] H. Lam. A Novel Method to Construct Taxonomy Electrical
Appliances Based on Load Signatures,. IEEE Transactions
on Consumer Electronics, 2007.
[8] A. Narayanan, V. Toubiana, S. Barocas, H. Nissenbaum, D.
Boneh: A Critical Look at Decentralized Personal Data
Architectures CoRR abs/1202.4503: (2012)
[9] H. Nissenbaum, Privacy in context: Technology, policy, and
the integrity of social life,”Stanford Law Books, 2010.
[10] J. Park and R. Sandhu, “The UCONABC usage control
model,” ACM Trans Inormationf System Security, vol. 7, no.
1, pp. 128-174, 2004.
[11] A. Pentland et al. Personal Data: The Emergence of a New
Asset Class. World Economic Forum. January 2011.
[12] S. Petronio, Unpacking the paradoxes of privacy in CMC
relationships: The challenges of blogging and relational
communication on the internet, In Computer-mediated
communication in Personal Relationships, 2011.
[13] The World Economic Forum. Rethinking Personal Data:
Strengthening Trust. May 2012.
[14] N. Zhang, W. Zhao: Distributed privacy preserving
information sharing. VLDB 2005.