Raffael Marty discusses using log visualization to detect insider threats. He outlines an insider detection process that involves building a list of precursor activities, assigning them scores, applying the precursors to log files, and visualizing results to surface insider candidates. Visualization helps analyze data access patterns, financial transactions, and tune the detection process by grouping similar user behaviors. Improvements include bucketizing precursors and using watch lists to adjust user scores.

1. Insider Threat Visualization
Raffael Marty, GCIA, CISSP
Chief Security Strategist @ Splunk>
Hack In The Box - September 07 - Malaysia

2. Who Am I?
Chief Security Strategist and Product Manager @ Splunk>
Manager Solutions @ ArcSight, Inc.
Intrusion Detection Research @ IBM Research
http://thor.cryptojail.net
IT Security Consultant @ PriceWaterhouse Coopers Applied
Security
Open Vulnerability and Assessment Language (OVAL) board Visualization
Common Event Expression (CEE) founding member 2008
Passion for Visualization
http://secviz.org
http://afterglow.sourceforge.net
2

3. Agenda
Convicted
Goal:
Visualization
Log Data Processing
Insider Detection Using
Data to Graph Visualization
AfterGlow and Splunk
Insider Threat
Insider Detection Process
Precursors
Scoring
Watch Lists
3

4. It’s Not That Easy
4

5. Convicted
In February of 2007 a fairly large information leak
case made the news. The scientist Gary Min faces up
to 10 years in prison for stealing 16,706 documents
and over 22,000 scientific abstracts from his
employer DuPont. The intellectual property he was
about to leak to a DuPont competitor, Victrex, was
assessed to be worth $400 million. There is no
evidence Gary actually turned the documents over to
Victrex.
5

6. DuPont Case
How It Could Have Been Prevented
What’s the answer?
6

7. DuPont Case
Log Collection!

8. DuPont Case
Simple Solution
8

9. DuPont Case
More Generic Solution
user
server
9

10. Visualization Questions
• Who analyzes logs?
• Who uses visualization for log analysis?
• Who is using AfterGlow?
• Have you heard of SecViz.org?
• What tools are you using for log
analysis?
10

11. Visualization
Answer questions you
didn’t even know of
✓ Quickly understand thousands of data entries
Increase Efficiency ✓ Facilitate communication
✓ Increase response time through improved
understanding
Make Informed Decisions
11

12. Insider Threat Visualization
• Huge amounts of data
• More and other data sources than for the traditional security use-cases
• Insiders often have legitimate access to machines and data. You need to log
more than the exceptions
• Insider crimes are often executed on the application layer. You need
transaction data and chatty application logs
• The questions are not known in advance!
• Visualization provokes questions and helps find answers
• Dynamic nature of fraud
• Problem for static algorithms
• Bandits quickly adapt to fixed threshold-based detection systems
• Looking for any unusual patterns
12

13. Visualizing Log Data
Parsing
Jun 17 09:42:30 rmarty ifup: Determining IP information for eth0...
Jun 17 09:42:35 rmarty ifup: failed; no link present. Check cable?
Jun 17 09:42:35 rmarty network: Bringing up interface eth0: failed
Jun 17 09:42:38 rmarty sendmail: sendmail shutdown succeeded
Jun 17 09:42:38 rmarty sendmail: sm-client shutdown succeeded
Visual
Jun 17 09:42:39 rmarty sendmail: sendmail startup succeeded
Jun 17 09:42:39 rmarty sendmail: sm-client startup succeeded
Jun 17 09:43:39 rmarty vmnet-dhcpd: DHCPINFORM from 172.16.48.128
Jun 17 09:45:42 rmarty last message repeated 2 times
Jun 17 09:45:47 rmarty vmnet-dhcpd: DHCPINFORM from 172.16.48.128
Jun 17 09:56:02 rmarty vmnet-dhcpd: DHCPDISCOVER from 00:0c:29:b7:b2:47 via vmnet8
Jun 17 09:56:03 rmarty vmnet-dhcpd: DHCPOFFER on 172.16.48.128 to 00:0c:29:b7:b2:47 via vmnet8
NH
✓ Interpret Data
✓ Know Data Formats
✓ Re-use don’t re-invent
✓ Find parsers at: http://secviz.org/?q=node/8
13

14. Charts - Going Beyond Excel
• Multi-variate graphs 10.0.0.1
10.12.0.2
- Link Graphs
UDP TCP
- TreeMaps HTTP
DNS
UDP TCP
- Parallel Coordinates SSH
SNMP FTP
14

15. Beyond The Boring Defaults For Link Graphs
10.0.0.1
SIP Name DIP
10.12.0.2
15

16. Link Graph Shake Up
[**] [1:1923:2] RPC portmap UDP proxy attempt [**]
[Classification: Decode of an RPC Query] [Priority: 2]
06/04-15:56:28.219753 192.168.10.90:32859 -> 192.168.10.255:111
UDP TTL:64 TOS:0x0 ID:0 IpLen:20 DgmLen:148 DF
Len: 120
SIP Name DIP SIP DIP DPort
192.168.10.90 portmap 192.168.10.255 192.168.10.90 192.168.10.255 111
SIP SPort DPort Name SIP DIP
192.168.10.90 32859 111 RPC portmap 192.168.10.90 192.168.10.255
16

17. TreeMaps ?
UDP TCP
HTTP
DNS What is this?
UDP All Network Traffic
TCP
SSH
SNMP FTP
17

18. TreeMaps Explained Treemap2 (http://www.cs.umd.edu/hcil/treemap)
20% 80%
UDP TCP
HTTP Size: Count
DNS
UDP TCP Color: Service
SSH
SNMP FTP
Configuration Hierarchy: Protocol -> Service
18

19. What’s Splunk?
1. Universal Real Time Indexing
2. Ad-hoc Search & Navigation search navigate alert report share
3. Distributed / Federate Search
4. Interactive Alerting & Reporting IT Search Engine
The
5. Knowledge Capture & Sharing
Router
Firewall
logs configurations scripts & code messages
Switch
Web Server
App Server
traps & alerts activity reports Database stack traces metrics
19

20. AfterGlow http://afterglow.sourceforge.net
Parser AfterGlow Grapher
Graph
CSV File LanguageFile
digraph structs {
graph [label="AfterGlow 1.5.8", fontsize=8];
node [shape=ellipse, style=filled,
fontsize=10, width=1, height=1,
aaelenes,Printing Resume fixedsize=true];
abbe,Information Encrytion edge [len=1.6];
aanna,Patent Access
aatharuy,Ping "aaelenes" -> "Printing Resume" ;
"abbe" -> "Information Encryption" ;
"aanna" -> "Patent Access" ;
"aatharuv" -> "Ping" ;
}
20

21. Why AfterGlow?
# Variable and Color
• Translates CSV into graph description
variable=@violation=("Backdoor Access", "HackerTool
Download”);
color.target="orange" if (grep(/$fields[1]/,@violation));
• Define node and edge attributes color.target="palegreen"
- color # Node Size and Threshold
- size maxnodesize=1;
size.source=$fields[2]
- shape size=0.5
sum.target=0;
• Filter and process data entries threshold.source=14;
- threshold filter Fan Out: 3 # Color and Cluster
color.source="palegreen" if ($fields[0] =~ /^111/)
- fan-out filter color.source="red"
color.target="palegreen"
- clustering cluster.source=regex_replace("(d+).d+")."/8"
21

22. AfterGlow - Splunk
Demo
./splunk <command>
./splunk search “<search command>” -admin <user>:<pass>
./splunk search "ipfw | fields + SourceAddress DestinationAddress" -auth
admin:changeme | awk ‘{printf”%s,%sn”,$1,$2}’ | afterglow -t -b 2 |
neato -Tgif -o test.gif
22

23. Insider Threat Definition
"Current or former employee or contractor who
• intentionally exceeded or misused an authorized level of
access to networks, systems or data in a manner that
• targeted a specific individual or affected the security of
the organization’s data, systems and/or daily business
operations"
[CERT: http://www.cert.org/insider_threat Definition of an Insider]
23

24. Three Types of Insider Threats
Information Theft is concerned
Fraud deals with the with stealing of confidential or
misuse of access proprietary information. This
privileges or the includes things like financial
intentional excess of Information statements, intellectual
access levels to obtain
Fraud property, design plans, source
Leak
property or services code, trade secrets, etc.
unjustly through
deception or trickery.
Sabotage
Sabotage has to do with any kind of
action to harm individuals,
organizations, organizational data,
systems, or business operations.
24

25. Insider Threat Detection
• Understand who is behind the crime
• Know what to look for
• Stop insiders before they become a problem
• Use precursors to monitor and profile users
• Define an insider detection process to
analyze precursor activity
25

26. Insider Detection Process
• Accessing job Web sites
• Build List of Precursors such as monster.com 1
• Assign Scores to Precursors • Sales person accessing
patent filings 10
• Printing files with "resume"
in the file name 5
• Sending emails to 50 or
more recipients outside of
the company
3
26

27. Insider Detection Process
Aug 31 15:57:23 [68] ram kCGErrorIllegalArgument:
• Build List of Precursors CGXGetWindowDepth: Invalid window -1
Aug 31 15:58:06 [68] cmd "loginwindow" (0x5c07)
set hot key operating mode to all disabled
• Assign Scores to Precursors Aug 31 15:58:06 [68] Hot key operating mode is now
all disabled
• Apply Precursors to Log Files Aug 27 10:21:39 ram com.apple.SecurityServer:
authinternal failed to authenticate user
raffaelmarty.
Aug 27 10:21:39 ram com.apple.SecurityServer:
Failed to authorize right system.login.tty by process /
usr/bin/su
do for authorization created by /usr/bin/sudo.
Apr 04 19:45:29 rmarty Privoxy(b65ddba0)
Request: www.google.com/search?q=password
+cracker
27

28. Insider Detection Process
• Build List of Precursors
• Assign Scores to Precursors
• Apply Precursors to Log Files
• Visualize Insider Candidate List
28

29. Insider Detection Process Engineer
• Build List of Precursors
• Assign Scores to Precursors
• Apply Precursors to Log Files
• Visualize Insider Candidate List
• Introduce User Roles
Legal
29

30. Insider Detection Process
?
• Build List of Precursors
• Assign Scores to Precursors
• Apply Precursors to Log Files
• Visualize Insider Candidate List
• Introduce User Roles
• Where Did the Scores Go?
30

31. Visualization for Insider Detection
• Visualization as a precursor
- analyze data access per user role
- find anomalies in financial transactions
• Documentation and communication of activity
• Tuning and analyzing process output
- groups of users with similar behavior
- groups of users with similar scores
31

32. Process Improvements
• Bucketizing precursors:
- Minimal or no impact
- Potential setup for insider crime
- Malicious activity okay for some user roles
- Malicious activity should never happen
- Insider Act
• Maximum of 20 points per bucket
• Using watch lists to boost / decrease scores for specific groups of
users
- Input from other departments (HR, etc.)
32

33. Tiers of Insiders
Nothing to On a bad track of Very likely Malicious
worry about just going malicious has malicious Insiders
yet intentions
0 20 60 80 100
33

34. The Insider? Finally?
34

35. Summary
• Log visualization
• Beyond the boring chart defaults
• AfterGlow and Splunk
- The free way to understanding your data
• Insider threat
• Insider detection process
35

36. Thank You
www.secviz.org
raffael.marty@splunk.com
raffy.ch/blog