Developing Analytic Technique and Defeating Cognitive Bias in Security

Defeating Cognitive Bias and

Developing Analytic Technique

Chris SandersBSides Augusta 2014

Chris Sanders

• Christian & Husband• Kentuckian and South

Carolinian• MS, GSE, CISSP, et al.• Non-Profit Director• BBQ Pit Master

Chris Sanders

Chris Sanders

“[Practical Packet Analysis] gives you everything you need, step by step, to become proficient in packet analysis. I could not find a better book.”

– Amazon Reviewer

Outline

Objectives: What is Analysis? What is Bias? Recognizing Bias Defeating Bias Analysis Methods

“How to make better technical decisions in any kind of security analysis.“

**Disclaimer**

I’m going to talk about matters of the brain, not sure the normal tech stuff.

My research for this presentation involved consultation with psychologists.

I, however, am not one.

Bias – A very personal story

2 AM

The Pain Begins

*Dramatization

Ultrasounds == Magic?

At this point…

So, I went to see a surgeon…

“Let’s Cut it Out!” - Surgeon

Missing Parts

Thus…

“Would it be accurate to say that I’m a medical miracle?” - Me

“Absolutely.” – Surgeon

Cause and Effect

• Cause: Bias…lots of it!– Confirmation Bias– Outcome Bias– Congruence Bias

• Effect: Unnecessary Surgery– 1 Week Recovery– Financial Loss– Pessimism Bias

Analysis

Analysis is Everywhere

• Making judgments based upon data• Security Analysis Happens for:– Malware Analysts– Intelligence Analysts– Incident Response Analysts– Forensic Analysts– Programming Logic Analysts

• My main focus is network intrusion analysis, so this talk will be framed through that.

Network Security Monitoring

• The collection, detection, and analysis of network security data.

• The goal of NSM is escalation, or to declare that an incident has occurred to that incident response can occur.

Evolution of NSM Emphasis

The Need for Analytic Technique

• Kansas State University Anthropological Study on SOCs - Key Finding:– “SOC analysts often perform sophisticated

investigations where the process required to connect the dots is unclear even to analysts.”

• Analysis == “Tacit Knowledge”

Analysis: Thinking About Thinking

• We need to critically examine how we think about information security analysis.

• We aren’t alone!– Scientific– Medical– Legal

Perception vs. Reality

• Perception: – “A way of regarding, understanding, or interpreting

something.”

• Reality: – “The state of things as they actually exist.”

Let’s take a test…

RED

GREEN

BLUE

BLACK

YELLOW

Test Results

• Variation of Stroop Test (John Stroop, 1935)• Measures Cognition – The Process of Perception

• Identifies Gap Between Perception & Reality• Used to Measure– Selective Attention– Cognitive Flexibility– Processing Speed

What is Bias?

“Prejudice in favor of or against one thing, person, or group compared with another, usually in a way considered to be unfair.”

•Perception != Reality•Perception is Everything, but Fallible•We tend to perceive what we expect/are conditioned to perceive

I’m Going to Show You an Image

I’m Going to Show You a Picture of a White Vase.

First Image Results

• Prompted for Face– 88% See Face– 12% See Sax Player

• No Prompt– 57% See Face– 43% See Sax Player

Second Image Results

• Prompted for Vase– 94% See White Vase– 6% See Two People

• No Prompt– 62% See White Vase– 38% See Two People

Bias Examples

Let’s Hit Closer to Home…

A Recent Example

Anchoring

• Defined: Heavily relying on a single piece of information.

• Examples:– Src/Dst Country -> OMG China!– IDS Alert Name -> It say this is X, so it must be X.– Timing -> It’s every 5 minutes!

Clustering Illusion

• Defined: Overestimating the value of perceived patterns in random data.

• Examples:– The great “beaconing”

fallacy– Unguided Visualizations

Availability Cascade

• Defined: Strong belief in something due to its repetition in public discourse

• Example:– “Chinese Traffic is Bad.”– “That rule generates a lot of false positives.”

Belief Bias

• Defined: Occurs when a decision is based on the believability of the conclusion.

• Examples:– “We wouldn’t be a target for a nation-state actor.”– “This is probably a false positive because it’s

unlikely someone would attack our VoIP system.”

Confirmation Bias

• Defined: Interpreting data during analysis with a focus on confirming one’s preconception.

• Ego is a big factor here

• Examples:– “I think this is nothing.”– “I think there is something going on here.”

Impact Bias

• Defined: Tendency to overestimate the significance of something based on the potential impact.

• Signature/Alert Naming + Lack of Experience Contribute to this.

• Example:– “The alert says this is a known APT1 back door, so I

need to spend all day looking at this.”

Irrational Escalation

• Defined: Justifying increased time investment based on existing time investment when it may not make sense.

• Sunk Cost Fallacy• Example:– “What do you mean this is nothing? I’ve spent all

day looking at this. I’ll spend all day tomorrow digging into it; I’m sure I’ll find something else there.”

Framing Effect

• Defined: Interpreting information differently based on how or from whom it was presented.

• Important in interaction with other analysts• Example:– Old Vet: “Steve doesn’t know what he is doing, so if

he is telling me this it probably doesn’t mean much.”– New Guy: “None of the more experienced guys said

anything about this, so it must not matter.”

Overconfidence Effect

• Defined: Excessive confidence in ones own decisions, especially in light of contrasting data.

• Example: • 99% Paradox – “I’m 99% sure this is right.”• One psych study suggest this statement is

wrong ~40% of the time.

Pro-Innovation Bias

• Defined: Excessive optimism and biased decisions based on an invention of one’s own making being involved in the analysis.

• Invention == System / Code / Concept• Example:– “My tool can do that.”– “I wrote that signature so I know it’s accurate.”– “This fits perfectly in my model!”

There are over 100 types of bias. How can we overcome them?

Overcoming Bias

What Can We Do?

• Preconception and Bias Cannot Be Fully Avoided

• Therefore: – Develop Repeatable Analytic Technique– Recognize Key Assumptions– Allow them to be Challenged

Analytic Techniques

Common Techniques:– Relational

Investigation– Differential

Diagnosis

Relational Investigation

• “Link Analysis”• Commonly Used in Criminal Investigations• Focuses on Entities, Relationships,

Interactions, and Degrees of Separation

Relational Investigation

Setting the Stage – Primary Relationships

Partial Story – Secondary Relationships

Full Attack Diagram – Tertiary Relationships

Differential Diagnosis

• Commonly Used in Medical Diagnosis

• Relies on Lists of Possibilities, and Systematically Eliminating Possibilities

Differential Diagnosis

Incident M&M

• Dr. Ernest Codman at Mass. General Hospital• Post-Patient Meetings to Discuss What

Occurred and How to Better It• Incident M&M

1. Handler/Analyst Presents Case2. Followed by Alternative Analysis

Alternative Analysis

• Developed by Richards Heuer Jr. (FBI)• Series of Peer Analysis Methods• Designed to Help Overcome Bias and Improve

Quality of Analysis

Group A / Group B

• Group A – Presenting Analyst/Team• Group B – Secondary Analyst/Team

• Two Independent Analysis Efforts• Note are Compared During the Presentation• Identify Differing Conclusions from Same Data

Red Cell Analysis

• Peer Focus on Attacker’s Viewpoint• Questioning in Relation to Attackers Perceived

Goals• Requires Some Offensive Experience• Best Executed by Red Team if Available

What If Analysis

• Focus on Cause/Effect of Actions That May Not Have Actually Occurred– What is the attacker had done X? How would you

have changed your approach?– What if you didn’t stumble across X in Y data?

• Enhances Later Investigations

Key Assumptions Check

• Presenter Identifies Assumptions During Analysis

• Peers Challenge Assumptions• Pairs Well with “What If” Analysis– “What if it were possible for that malware to

escape that virtual machine?”– “Would you come to the same conclusion if you

knew this was APT3 instead of APT1?”

Incident M&M Best Practices

• Limit Frequency• Set Expectations• Require a Strong Mediator• Keep it at the Team Level – No Sr. Managers• Encourage Servant Leadership• Discourage Personal Attacks• Write it Down!

Conclusion

• The Era of Analysis is Upon Us• Bias is Inevitable – Learn to Recognize It• Overcome Analysis Hurdles With:– Analytic Technique– Alternative Analysis

Thank You!

E-Mail: [email protected]: @chrissanders88

Blog: http://www.chrissanders.orgBook Blog: http://www.appliednsm.com

Testimony: http://www.chrissanders.org/mytestimony