Cognitive Security:
How Artificial Intelligence Is Your New Best Friend
TM
The potential for machine learning in the cyber space
KEITH MOORE
DIRECTOR OF PRODUCT MANAGEMENT
SPARKCOGNITION
Why Machine Learning Is Needed To Solve These Problems
Automates the analyst
research process
Scales to ingest massive data
streams
Combats constantly evolving
malware variants
Defends networks against hard
to identify APTs
Cross-correlates between data
to find threats
SparkCognition A.I. technology can accelerate Decision Making
• Identifies anomalous events
• Aggregates multiple data streams
• Recognizes known and unknown
patterns
• Incorporates analyst feedback so that
underlying models learn from human
response
• Presents actionable evidence behind its
conclusions
A.I software trains on historical events to recognize patterns and provide maximum business awareness
Scan for matches Against DB and
Suspected Patterns
Patterns Stored in
Cognitive DB
Supervisory Input
Confidential
TM
What sort of problems can be solved using machine learning?
Polymorphic malware is significantly shifting the security landscape
 78% of security analysts no longer trust anti-virus tools
 99% of malware hashes are seen for only 58 seconds or less
 16% of malware samples are “virtual machine aware”
Machine Learning Anti-Virus combats obfuscation and
polymorphism
Break down the
DNA of every file
Analyze all of the
components
individually
Determine
likelihood of
malicious nature
• 50% of analysts cite too many false
positives as a significant detractor of
SIEM use
SIEM
Big data is leading to a big problem…
10,000 Alerts
• Analysts can focus on real threats with
much of their research completely
automated
SIEM
Machine Learning research and prioritization tools ensure
analysts look at relevant threats
10,000 Alerts
Identifying terms are
pulled from potential
threat anomalies
Multiple search engines are
automatically queried (e.g.: “Is
Opera/ 12.14 using Port 8888 a
threat?” )
Search engine results
are filtered for
language and
relevance
Threat Term Filter
Threat Confidence
& Evidence
NLP Model
Processing
Summary
Generation
Search engine
results are
aggregated
Proprietary NLP model reads
and understands language,
assigns confidence score
reflecting malicious nature
Extraction
Search
Engine 2
Search
Engine 1
Aggregate
Results
Relevant term text
is extracted from
web pages
Most relevant
term text is
identified and
ranked
Evidence is summarized
using natural language
generation and displayed
with confidence score
Search
Engine 3…
Natural Language Processing builds a bridge between anomalous
behavior and malicious intent
SparkSecure is a comprehensive, advanced cyber security platform
Agentless EP
Protection
Bot Detection Find the
Snowden
Personally
Identifiable Info
Web Server
Protection
Research
Automation
• Traditional AV detects
< 5% of new
advanced threats
• 56% of web traffic is bot
generated
• 29% of bot traffic is
malicious
• 11% of employees
access unauthorized
docs and sell for profit
• Companies need to
prevent the leakage of
PII. Out of compliance
can lead to penalties
• Web server breaches,
on average, cost $3.79M
• Analysts are inundated
with alerts, most of
which are false positives
• Forensic costs went up
25% last year
• Ingests network traffic
logs to monitors
network perimeter for
anomalies
• Deploys Machine
Learning AntiVirus to
detect 98% of new
zero-day attacks early
• Proprietary Machine
Learning classification
algorithm powers bot
identification
• Develops Bot signatures
and rules to block
threats
• Uses temporal and
behavioral analysis to
identify deviations and
threats with minimal
false positives
• Automatically examine
user agent and payloads
for PII
• Stop inbound &
outbound leakage
• Reads email traffic and
attachments for
unstructured PII
• Analyzes incoming traffic
for SQL injections, XSS,
DDoS etc.
• Co-relates to multiple
internal & external
sources
• Automated threat
research expedites time
to remediation
• Rapid custom data
querying in HDFS scales
to massive data sets
• IBM Watson powered
automated threat
research and advisor
ProblemSolution
TM
Thank You

Cognitive Security: How Artificial Intelligence is Your New Best Friend

  • 1.
    Cognitive Security: How ArtificialIntelligence Is Your New Best Friend
  • 2.
    TM The potential formachine learning in the cyber space KEITH MOORE DIRECTOR OF PRODUCT MANAGEMENT SPARKCOGNITION
  • 3.
    Why Machine LearningIs Needed To Solve These Problems Automates the analyst research process Scales to ingest massive data streams Combats constantly evolving malware variants Defends networks against hard to identify APTs Cross-correlates between data to find threats
  • 4.
    SparkCognition A.I. technologycan accelerate Decision Making • Identifies anomalous events • Aggregates multiple data streams • Recognizes known and unknown patterns • Incorporates analyst feedback so that underlying models learn from human response • Presents actionable evidence behind its conclusions A.I software trains on historical events to recognize patterns and provide maximum business awareness Scan for matches Against DB and Suspected Patterns Patterns Stored in Cognitive DB Supervisory Input Confidential
  • 5.
    TM What sort ofproblems can be solved using machine learning?
  • 6.
    Polymorphic malware issignificantly shifting the security landscape  78% of security analysts no longer trust anti-virus tools  99% of malware hashes are seen for only 58 seconds or less  16% of malware samples are “virtual machine aware”
  • 7.
    Machine Learning Anti-Viruscombats obfuscation and polymorphism Break down the DNA of every file Analyze all of the components individually Determine likelihood of malicious nature
  • 8.
    • 50% ofanalysts cite too many false positives as a significant detractor of SIEM use SIEM Big data is leading to a big problem… 10,000 Alerts
  • 9.
    • Analysts canfocus on real threats with much of their research completely automated SIEM Machine Learning research and prioritization tools ensure analysts look at relevant threats 10,000 Alerts
  • 10.
    Identifying terms are pulledfrom potential threat anomalies Multiple search engines are automatically queried (e.g.: “Is Opera/ 12.14 using Port 8888 a threat?” ) Search engine results are filtered for language and relevance Threat Term Filter Threat Confidence & Evidence NLP Model Processing Summary Generation Search engine results are aggregated Proprietary NLP model reads and understands language, assigns confidence score reflecting malicious nature Extraction Search Engine 2 Search Engine 1 Aggregate Results Relevant term text is extracted from web pages Most relevant term text is identified and ranked Evidence is summarized using natural language generation and displayed with confidence score Search Engine 3… Natural Language Processing builds a bridge between anomalous behavior and malicious intent
  • 11.
    SparkSecure is acomprehensive, advanced cyber security platform Agentless EP Protection Bot Detection Find the Snowden Personally Identifiable Info Web Server Protection Research Automation • Traditional AV detects < 5% of new advanced threats • 56% of web traffic is bot generated • 29% of bot traffic is malicious • 11% of employees access unauthorized docs and sell for profit • Companies need to prevent the leakage of PII. Out of compliance can lead to penalties • Web server breaches, on average, cost $3.79M • Analysts are inundated with alerts, most of which are false positives • Forensic costs went up 25% last year • Ingests network traffic logs to monitors network perimeter for anomalies • Deploys Machine Learning AntiVirus to detect 98% of new zero-day attacks early • Proprietary Machine Learning classification algorithm powers bot identification • Develops Bot signatures and rules to block threats • Uses temporal and behavioral analysis to identify deviations and threats with minimal false positives • Automatically examine user agent and payloads for PII • Stop inbound & outbound leakage • Reads email traffic and attachments for unstructured PII • Analyzes incoming traffic for SQL injections, XSS, DDoS etc. • Co-relates to multiple internal & external sources • Automated threat research expedites time to remediation • Rapid custom data querying in HDFS scales to massive data sets • IBM Watson powered automated threat research and advisor ProblemSolution
  • 12.