Transcript Slides - Applied Crypto Group at Stanford University

Are “Trusted Systems” Useful for
Privacy Protection?
Joan Feigenbaum
http://www.cs.yale.edu/~jf
PORTIA Workshop
Stanford Univ., July 8-9, 2004
1
General Problem: It is Hard to Exercise
“Remote Control” over Sensitive Data
 Many machine-readable permissions
systems
– Rights-management languages
– Privacy policies
– Software licenses
 None is now technologically enforceable.
– All software-only permissions systems can be
circumvented.
– Once data are transferred, control is lost.
2
Proposed Solution: “Trusted Systems”
 We need definition(s) of “trusted systems.”
Key feature: hardware-based, cryptographic support for
proofs that remote machines are running approved
software stacks
 Useful feature
– Leads to interesting theory?
– Feasible to build and deploy?
 Solves practical problems?
3
Trusted Systems for the Distribution of
Entertainment Content
 Content providers’ problem is piracy. Approved
software stacks would implement “draconian DRM.”
– No unauthorized use of content
– No unauthorized access to or transfer of content
 Unclear whether these goals, especially the second,
can be achieved. See, e.g.,
– Biddle et al. 2002: “Darknet” creates unauthorized access and
transfer.
– Haber et al. 2003: Technological weaknesses
 If these goals were achieved, the piracy problem
would be (at least partly) solved.
4
Dan Geer at YLS: “DRM ≡ Privacy”
This claim is made often and disputed often in
the trusted-system context. Is it true?
 What (and whose) privacy problem is being addressed?
 What is the approved software stack required to do?
 Are these technical requirements achievable?
 If they were achieved, would the privacy problem be
solved?
5
Example: Privacy of Medical Data
Whose problem?
 Individual data subject
– Often, he or she is not the entity that creates the data
or the one that must demand proof from a potential
data recipient.
– Different from DRM for entertainment content.
 Institution
Hospital, drs’ office, insurance co., etc.
6
Privacy of Medical Data, cont.
What problem?
Example: Compliance with HIPAA rules about when
personal medical data may be disclosed.
 Targeted attack on specific data subject.
Trusted systems won’t prevent this disclosure!
– Small data objects (some only one bit) abound. Attacker can
disclose them without computers. General fact about trusted
systems (but not applicable to DRM for entertainment)
– Data objects obtainable by other means. General fact (sometimes
applicable to DRM for entertainment, as in “Darknet” )
 General disregard for or misunderstanding of rules
7
General (Untargeted) HIPAA-Compliant Privacy
 Medical-privacy logic is incomplete. Complying with
it requires good human judgment.
– “Life-threatening situation”
– “Public-health emergency”
– “Poses a threat to himself or others”
Trusted systems can’t fully solve this problem!
 This is a general fact about trusted systems; they
can’t solve problems that are inherently unsolvable
by computers.
– Not necessarily relevant to draconian DRM
– Relevant to the more general problem of copyright compliance,
which includes fair use
8
Conclusion: Trusted Systems Cannot
Ensure Privacy of Personal Medical Data
 Circumvention of privacy-protection
mechanisms is not the most pressing threat
to medical-data privacy.
 More generally, trusted systems will be
ineffective when
– There are alternative ways to obtain the sensitive
data (or accomplish the attacker’s goals).
– The relevant policy is not conveniently expressible as
a program.
 When will trusted systems be effective?
9
Policy-Enforcement Mechanisms in MS
Products (BAL/cs155/03-25-03)
 MS DRM for eBooks
 MS DRM for Windows
Media
 Windows Rights
Management Services
Office 2003 Information
Rights Management
 License servers for
 Ultimate TV/eHome (digital



Terminal Services, File &

Print Services, etc.
 Xbox (anti-repurposing)
storage of video)
File system ACLs
Enterprise policy
management
Group policy in domains
Partially trusted code
policies (.NET Framework)
NGSCB
10
Which Attributes are Relevant?





Number of data sources?
Number of data recipients?
Need to retransmit data?
Stability of policies?
Third-party certifiers?
•••
When could trusted systems help?
11
Whenever “The Program is the Policy?”
 Organizations mandate the use of particular
software environments for a variety of
reasons, some objectionable.
 Are there benign reasons to do so?
– Absence of particular serious flaws
– Mission-critical properties, e.g.,
• Maintains audit trails
• Does not maintain audit trails
– Protection against subversion and misrepresentation
(analogy with copy-left licenses)
 What else are trusted systems good for?
12