1. Andrei G. Stoica and Csilla Farkas
Integrated Security Framework for Semantically Enhanced Semi-Structured Data
Andrei G. Stoica and Csilla Farkas
Department of Computer Science & Engineering University of South Carolina i

2. Overview Machine understandable data semantics:
domain and context definition
ontologies
metadata
What are the security implications?
New security mechanisms?
New security paradigm?

3. XML Language High-level application messaging
Used for storage application
reduces computation overhead
uniform access
Base for semantic orientated languages - RDF, DAML
Increased popularity

4. Semantic Tools
The information process is augmented with a semantic layer.
Infrastructure allows computers
to reason about data meaning.
Computers exchange information
transparently on behalf of the user.
Implications
Intelligent high-volume processing

5. Security Setup
Increased Connectivity + Extensive XML support + Semantic Infrastructure = New Security Threats
Established Security Models do not address this dimension:
Indirect disclosure
Undesired Inference
Available inference models difficult to transfer from database security
open domains

6. Related Work Document Instance Security XML Access Control Models
Digital Signatures
Encryption
XML Access Control Models
Security labels assignment
Multi-level XML Security
Extensions from Database Security

7. Problems? Semantic correlations ignored Inconsistent reply
Indirect unauthorized disclosure

8. Example View over UC data medicalFiles <medicalFiles> UC
<countyRec> S
<patient> S
<name>John Smith </name> UC
<phone> </phone> S
</patient>
<physician>Jim Dale </physician> UC
</countyRec>
<milBaseRec> TS
<name>Harry Green</name> UC
<phone> </phone> S
<physician>Joe White
</physician> UC
<milTag>MT78</milTag> TS
</milBaseRec>
</medicalFiles>
countyRec
milBaseRec
physician
Jim Dale
physician
Joe White
milTag
MT78
patient
patient
name
John Smith
phone
name
Harry Green
phone
View over UC data

9. Inference Set of data + associations derive the target data
Traditionally a human task
At the limit, infer any target
given enough related data
and metadata.

10. Problems? If the inference target is confidential information
Security Violation

11. Example Simulation Exploitation Using Open Source Information:
Objective: US Government would like to share a limited simulation software with friendly countries.
Can this software be used to explore the capabilities of US weaponry?
Can sufficient information be found from public sources to create such simulation?

12. Example
Findings:
Most of the information needed for the simulation was available on the Internet.
Needed human aid to combine available information

13. Proposed Solution What do we do?
XML Views Considering Semantic Dimension.
do not disclose more information (including structure of the document).
cover stories.
Web Inference
make sure the information we publish does not lead to our confidential data.

14. Proposed Solution XML Access Control Global Disclosure Control
semantic consistent reply
prevent illegal inference from
query reply (cover stories).
Global Disclosure Control
detect and prevent a set of undesired inferences using public Internet data
in correlation with public local data

15. Global Data Privacy Control
Security Engine
Local Organization
Access Control
Corrective Measures
Request
SecView
Local
XML
Database
Interface
Module
Return
Oxsegin
Update
Local
Ontology
Upload
Global Data Privacy Control
Local
Data
Internet
Data

16. Secure XML Views
Builds secure & semantic consistent single security level partial views
Minimum Semantic Conflict Graph
avoids semantic conflicts
Multi-Plane DTD Graph MPG
structural relationships between tags
Andrei Stoica, Csilla Farkas. “Secure XML Views”, In Proc. of IFIP 2002

17. Example DTD Graph MSCG medicalFiles name phone countyRec milBaseRec
emrgRec
physician
patient
milTag
physician
name
phone

18. Oxsegin Inference Engine Security Violations Corrective Measures Local
Classified
Database
Inference Engine
Local
Public
Database
Security Violations
Internet
Databases
Corrective
Measures

19. Corrective Measures Local Public Data Remove information
Release misleading information
Internet Public Data
Target desirable inference results

20. Inference Engine Replicated Data Inf. Violation Prob. Coef. Pointers
Public+Local
Database
Local
Classified
Database
Violation
Pointers
Prob.
Coef.
Correlated
Data Inf.
Inf. Struct Ontology

21. Replicated Data Inference
Identifies replicated information under different security classifications
Violation Pointer = similar units of data at different security levels
Inference is guided by inference structures built on ontology concept hierarchy
Andrei Stoica, Csilla Farkas. “Ontology guided XML Security Engine”, In Journal of Intelligent Information Systems, to appear.

22. Replicated Data Inference
Inf. Tree
Ontology
Public
Data file
Classified
Data file  A
Patriot
Freq. N0 M1 B B C
,  M2 M4 M3 N1 N2 D E
PAC-2
Freq.
PAC-3
Freq.
PAC-2
Freq.
PAC-3
Freq. M7 M7 N5 N5 N6 N7
Scientific data on
radar components
Missiles Tracking Systems
Confidence Level (M7,N5) = ƒ (,,,)

23. Correlated Data Inference
Identifies sensitive data in the public domain (relative to a given classified database – usually the local database).
Inference guidance:
Ontology concept hierarchy
Structural similarity of public data
Csilla Farkas, Andrei Stoica. “Correlated Data Inference, Ontology Guided XML Security Engine”, In Proc of IFIP 2003.

24. Correlated Data Inference
Features of similarity:
Levels of abstraction for each node
Distance of associated nodes from association root
Similarity of the distances
Length of the distance
Similarity of sub-trees originating from correlated nodes

25. Correlated Data Inference
Association similarity:
Distance of each node from the association root
Difference of the distance of the nodes from the association root
Similarity of the sub-trees originating at nodes
Air show
address
fort
address
fort

26. Correlated Data Inference
Object[].
waterSource :: Object
basin :: waterSource
place :: Object
district :: place
address :: place
base :: Object
fort :: base
fort
address
basin
district ?
base
Water source

27. Correlated Data Inference
Object[].
waterSource :: Object
basin :: waterSource
place :: Object
district :: place
address :: place
base :: Object
fort :: base
place
base
Public
address
fort
Public
district
basin
Water Source
Water source
base
Confidential

28. Summary
Secure XML Views provide semantic consistent query reply and cover stories.
Oxegin architecture and methods detect undesired inferences
Structural similarity
Semantic concept hierarchy
Confidence in derived inferences

29. Next Class
Stream data