Quantitative Risk Analysis Sanjay Goel University at Albany, SUNY Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information.

Transcript Quantitative Risk Analysis Sanjay Goel University at Albany, SUNY Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information.

Quantitative Risk Analysis

Sanjay Goel University at Albany, SUNY

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Course Outline

> Unit 1: What is a Security Assessment? – Definitions and Nomenclature Unit 2: What kinds of threats exist?

– Malicious Threats (Viruses & Worms) and Unintentional Threats Unit 3: What kinds of threats exist? (cont’d) – Malicious Threats (Spoofing, Session Hijacking, Miscellaneous) Unit 4: How to perform security assessment?

– Risk Analysis: Qualitative Risk Analysis Unit 5: Remediation of risks?

– Risk Analysis: Quantitative Risk Analysis

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis

Outline for this unit

Module 1: Quantitative Risk Analysis and ALE Module 2: Risk Aggregation Module 3: Case Study Module 4: Cost Benefit Analysis and Regression Testing Module 5: Modeling Uncertainties

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Module 1

Quantitative Risk Analysis and ALE

• • • • • • •

Outline

What is Risk Analysis?

What is Quantitative Risk Analysis?

What are the steps involved?

How to determine the Likelihood of Exploitation?

How to determine Risk Exposure?

How to compute Annual Loss Expectancy (ALE)?

Examples – Gym Locker – Hard Drive Failure – Virus Attack

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Risk Analysis Definition

• Risk analysis involves the identification and assessment of the levels of risks calculated from the known values of assets and the levels of threats to, and vulnerabilities of, those assets.

• It involves the interaction of the following elements: – Assets – Vulnerabilities – Threats – Impacts – Likelihoods – Controls

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Risk Analysis Concept Map

• Threats exploit system vulnerabilities which expose system assets. • Security controls protect against threats by meeting security requirements established on the basis of asset values.

Source: Australian Standard Handbook of Information Security Risk Management – HB231-2000

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Quantitative Risk Analysis

• Quantitative risk analysis methods are based on statistical data and compute numerical values of risk • By quantifying risk, we can justify the benefits of spending money to implement controls.

• It involves three steps – Estimation of individual risks – Aggregation of risks – Identification of controls to mitigate risk

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Risk Analysis Steps

Security risks can be analyzed by the following steps: • Identify and determine the value of assets • Determine vulnerabilities • Estimate likelihood of exploitation – Compute frequency of each attack (with & w/o controls) using statistical data • Compute Annualized Loss Expectancy – Compute exposure of each asset given frequency of attacks • Survey applicable controls and their costs • Perform a cost-benefit analysis – Compare exposure with controls and without controls to determine the optimum control

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Determining Assets and Vulnerabilities

• Identification of Assets and Vulnerabilities is the same for both Qualitative and Quantitative Risk Analysis • The differences in both of these is in terms of valuation: – Qualitative Risk Analysis is more subjective and relative – Quantitative Risk Analysis is based on actual numerical costs and impacts.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Determine Likelihood of Exploitation

• Likelihood relates to the stringency of existing controls – i.e. likelihood that someone or something will evade controls • Several approaches to computing probability of an event – classical, frequency and subjective • Probabilities hard to compute using classical methods – Frequency can be computed by tracking failures that result in security breaches or create new vulnerabilities can be identified – e.g. operating systems can track hardware failures, failed login attempts, changes in the sizes of data files, etc.

• Difficult to obtain frequency of attacks using statistical data.Why?

– Data is difficult to obtain & often inaccurate • If automatic tracking is not feasible, expert judgment is used to determine frequency

Sanjay Goel, School of Business/Center for Information Forensics and Assurance

University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Approaches

• Delphi Approach

– Probability in terms of integers (e.g. 1-10)

• Normalized

– Probability in between 0 (not possible) and 1 (certain)

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Delphi Approach Frequency

More than once a day Once a day Once every three days Once a week Once in two weeks Once a month Once every four months Once a year Once every three years 2 Less than once in three years 1 4 3 7 6 5 9 8

Ratings

10 • Subjective probability technique originally devised to deal with public policy decisions • Assumes experts can make informed decisions • Results from several experts analyzed • Estimates are revised until consensus is reached among experts

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Risk Exposure

• Risk is usually measured as $ per annum and is quantified by risk exposure.

– ALE (Annual Loss Expectancy, expressed as: $/year) • If an event is associated with a loss – LOSS = RISK IMPACT ($) • The probability of an occurrence is in the range of: – 0 (not possible) and 1 (certain) • Quantifying the effects of a risk by multiplying risk impact by risk probability yields risk exposure.

– RISK EXPOSURE = RISK IMPACT x RISK PROBABILITY

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Intangible Assets

• Incorporating intangible assets within Quantitative Risk Analysis is difficult as it is hard to put a price on things such as trust, reputation, or human life.

• However, it is necessary to put an as accurate a value as possible when factoring these assets within risk analysis as they may be even more important than tangible assets.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Computing ALE

• Single Loss Expectancy: Loss to an asset if event occurs – Value of the lost asset = Ci – Impact on the Asset (if event occurs) = Pi – SLE = Ci * Pi • Annualized Rate of Occurrence (ARO) characterizes, on an annualized basis, the frequency with which a threat is expected to occur.

• Annualized Loss Expectancy (ALE) computes risk using the probability of an event occurring over one year. • Formulation – ALE = (SLE)(ARO) • Source: Handbook of Information Security Management, Micki Krause and Harold F. Tipton

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Example #1: Gym Locker

Scenario: There is a gym locker used by its members to store clothes and other valuables. The lockers cannot be locked, but locks can be purchased.

You need to determine:

1) Risk exposure for gym members 2) Controls to reduce risk

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Example #1: Gym Locker, cont’d.

• Identify assets and determine value – Clothes $50 – Wallet – Glasses – Sports equipment $100 $100 $30 – Driver’s license – Car keys $20 $100 – House keys $60 – Tapes and walkman $40 ____ – Total Loss/week: $500 • Find vulnerability – Theft – Accidental loss – Disclosure of information (e.g. read wallet) – Vandalism

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Example #1: Gym Locker, cont’d.

• Estimate likelihood of exploitation – 10 (more than once a day) – 9(once a day) – 7 (once a week) – 6 (once every two weeks) – 5 (once a month) – 4 (once every four months) – 3 (once a year) – 2 (once every three years) – 1 (less than once every 3 years) • For theft: estimated likelihood is 7 • Figure annual loss: – ~$500 worth of loss each week, ~52 weeks in a year – ~$26,000 loss per year

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Example #1: Gym Locker, cont’d.

• Determine cost of added security – New lock $5 – Replacement for lost key $10 – On average members lose one key twice a month (24 times per year) • Estimate likelihood of exploitation under added security – The new likelihood of theft could be estimated at a 4. • Cost Benefit Analysis – Revised Losses (including cost of controls) = (500 * 4) + (15*24) = 2360 – Net savings = 26000 – 2360 = 23640

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Example #2: Hard Drive Failure

• The chance of your hard drive failing is once every three years – Probability = 1/3 • Intrinsic Cost – $300 to buy new disk • Hours of effort to reload OS and software – 10 hours • Hours to re-key assignments from last backup – 4 hours • Pay per hour of effort – $10.00 per hour • Total loss (risk impact) – $300 + 10 x (10+4) = $440 • Annual Loss Expectancy (pa = per annum) – (440 x 1/3)$pa = $147 pa

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Example #3: Virus Attack

• Situation: Virus Attack on same system – You frequently swap files with other people, but have no anti-virus software running.

– Assume an attack every 6 months (Probability = 2 per year) – No need to buy a new disk – Rebuild effort (10 + 4) hours – Total loss = $10 x (10 + 4) = $140 – ALE = ($140 x 2) $pa = $280 pa

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Questions 1 and 2

1) Why is it important to quantify risk?

2) Give the definitions for: a.

Single Loss Expectancy b. Annualized Rate of Occurrence c. Annual Loss Expectancy

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Risk Analysis and ALE

Question 3

3) For this situation: a.

Same system as examples 2 and 3

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Module 2

Risk Aggregation

• • •

Risk Aggregation

Outline

How do you determine risk posture?

What is this risk aggregation model?

Matrices – Asset/Vulnerability – Vulnerability/Threat – Threat/Control

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Risk Aggregation

Risk Posture

• Individual risks aggregated = Total risk posture – True comparison of relative risks of different organizations • Mathematical approach for aggregation provided – Methodology standardized – Data needs to be customized to organization • Controls can reduce the cost of exposure – Need to determine optimum controls for organization – Methodology for determining controls shown next slide • Analysis should be undertaken to see the impact of new projects on security

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Risk Aggregation

Model

• Let: – A be a vector of loss of an asset where a l – V be a vector of vulnerabilities where v k is the l th is the k th asset, s.t., 0 < l < L vulnerability, s.t., 0 < k < K – T be a vector of threats where t j is the j th – C be the vector of vulnerabilities where c i asset, s.t., 0 < j < J is the i th control, s.t., 0 < i < I – Also M α be the matrix that defines the impact of vulnerabilities (breach in security) on assets, where, α kl is the impact of k th vulnerability on the l th asset – Also M β where, β jk be the matrix that defines the impact of threats on the vulnerabilities, is the impact of jth threat on kth vulnerability – Also M γ be the matrix that defines the impact of a controls (breach in security) on the threats, where, γ ij is the impact of i th control on the j th threat The notation is graphically explained in the next few slides

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

•

Risk Aggregation

Model, cont’d.

A (Assets) a kl L K Where a kl is the Impact of vulnerability k on given asset l.

– i.e. fraction of the asset value that will be lost if the vulnerability is exploited

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

• Data Collection: – Primary Data from corporations that track financial losses due to different attacks – Secondary Data from the reports of financial loss from organizations like CERT, CSI/FBI and AIG – Data specific to a corporation, could perhaps be classified into different groups of companies

Risk Aggregation

Model, cont’d.

V (Vulnerabilities) b J jk K b jk is the probability that threat j will exploit vulnerability k • Data Collection: – Threat data and frequency of threats is information that is routinely collected in CERT and other such agencies.

– Log data and collected data from the organization itself can be another source of information – Data can also be collected via use of automated monitoring tools

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Risk Aggregation

Model, cont’d.

T (Threats) g ij J I g ij is the fraction by which controls reduce the frequency of a threat exploiting a vulnerability • Data Collection: – Approximate control data can be procured from various industry vendors who have done extensive testing with tools.

– Other sources of data can be independent agencies which do analysis on tools.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Risk Aggregation

Model, cont’d.

Then losses if no control exist



j J

k K L

    

t j

 b

 a



a l

Then losses if controls exist

* 

j J K

    1

l L

(

   1

 1 

)  b

 a



a l



i I

  1

C i



 ( 1  g

)  = sum  = product

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Risk Aggregation

Optimization

If ζ is the maximum allocated budget for controls the optimization problem can be formulated as:

Minimize

R

* 

j J

k K L I

  

  

 1 

)  b

 a



a

l where

i I

  1

C i

 

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Risk Aggregation

Question 1

1) How would you collect data for the following: a.

Assets and Values Potential Threats Exploitable Vulnerabilities Possible Controls

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Module 3

Case Study

• • •

Case Study

Outline

What is the case about?

What would fit into the categories of: – Assets – Vulnerabilities – Threats – Controls Filling in the matrices – Asset/Vulnerability – Vulnerability/Threat – Threat/Control

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Example

• Use the information that you have learned in the lecture in the following case study of a government organization. • Remember these key steps for determining ALE – Identify and determine the value of assets – Determine vulnerabilities – Estimate likelihood of exploitation – Compute ALE – Survey applicable controls and their costs – Perform a cost-benefit analysis

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Case

An organization delivers service throughout New York State. As part of the planning process to prepare the annual budget, the Commissioner has asked the Information Technology Director to perform a risk analysis to determine the organization’s vulnerability to threats against its information assets, and to determine the appropriate level of expenditures to protect against these vulnerabilities.

The organization consists of 4,000 employees working in 200 locations, which are organized into 10 regions. The average rate of pay for the employees is $20/hr. Cost benefit analysis has been done on the IT resource deployment, and the current structure is the most beneficial to the organization, so all security recommendations should be based on the current asset deployment.

Each of the 200 locations has approximately 20 employees using an equal number of desktop and laptop computers for their fieldwork. These computers are used to collect information related to the people served by the organization, including personally identifying information. Half of each employee’s time is spent collecting information from the clients using shared laptop computers, and half is spent processing the client information at the field office using desktop computers. Replacement cost for the laptops is $2,500 and for the desktop is $1,500.

Each of the 10 regions has a network server, which stores all of the work activities of the employees in that region. Each server will cost $30,000 to replace, plus 80 hours of staff time. Each incident involving a server costs the organization approximately $1,600 in IT staff resources for recovery. Each incident where financial records or personal information is compromised costs the organization $15,000 in lawyers time and settlement payouts. Assume that the total assets of the organization are worth 10 million dollars.

The organization has begun charging fees for the public records it collects. This information is sold from the organization website at headquarters, via credit card transactions. All of the regional computers are linked to the headquarters via an internal network, and the headquarters has one connection to the Internet. The headquarters servers query the regional servers to fulfill the transactions. The fees collected are approximately $10,000 per day distributed equally from each region, and the transactions are uniformly spread out over a 24 hour period.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Example- Assets (Tangible)

• Transaction Revenue- amount of profit from transactions • Data- client information • Laptops- shared, used for collecting information • Desktops- shared, used for processing client information • Regional Servers- stores all work activities of employees in region • HQ Server- query regional servers to fulfill transactions

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Example- Asset Valuations (Cost per Day) Transaction Revenue Data (Liability) Laptops Desktops Regional Servers HQ Server

$10,000 per day $10 million (total assets of organization) ½ x 200 (locations) x 20 (employees) x $2,500 (laptop cost) = $5,000,000 ½ x 200 (locations) x 20 (employees) x $1,500 (desktop cost) = $3,000,000 $30,000 (server cost)x 10 (regions) + 80 (hours) x $20 (pay rate) x 10 (regions)+ $10,000 (transaction revenue) = $326,000 $10,000 (transaction revenue) + $100,000 (cost of HQ server) + 80 (hours) x $20 (pay rate) x 10 (regions) = $126,000

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Example- Vulnerabilities

• Vulnerabilities are weaknesses that can be exploited • Vulnerabilities – Laptop Computers – Desktop Computers – Regional Servers – HQ server – Network Infrastructure – Software • Computers and Servers are vulnerable to network attacks such as viruses/worms, intrusion & hardware failures • Laptops are especially vulnerable to theft

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Example- Threats

• Threats are malicious & benign events that can exploit vulnerabilities • Several Threats exist – Hardware Failure – Software Failure – Theft – Denial of Service – Viruses/Worms – Insider Attacks – Intrusion and Theft of Information

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Example- Controls

•Intrusion detection and firewall upgrades on HQ Server – mitigate HQ server failure and recovery •Anti-Virus Software – mitigates threat of worms, viruses, DOS attacks, and some intrusions • Firewall upgrades – mitigates threats of DOS attacks and some intrusions, worms and viruses • Redundant HQ Server – reduces loss of transaction revenue •Spare laptop computers at each location – reduces loss of transaction revenue and productivity • Warranties – reduces loss of transaction revenue and cost of procuring replacements • Insurance – offset cost of liability • Physical Controls – reduce probability of theft • Security Policy – can be used to reduce most threats.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Asset/Vulnerability Matrix

• The coefficients of this matrix are usually based on internal data as well as financial loss organizations • For the current example we will assume data for illustration of the concept – Transactions are mostly associated with the regional servers which store the data, the HQ server which takes all requests, and the network infrastructure with which clients access the data. (.30 each) – Laptops, desktops and software is only associated with the remaining 10% (.033 each) – Data that is located on laptops and desktops make up only 10% of total data because they are only used for collecting and processing.

– The regional servers contain all other data.

– Other assets are associated at 100% with their respective vulnerabilities. (e.g. laptops with laptops, desktops with desktops, etc.)

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Asset/Vulnerability Matrix, cont’d.

Assets Transaction Revenue Data (Liability) Laptops Desktops Vulnerabilities

Input Asset Values

 10,000 10,000,000 5,000,000 3,000,000 Regional Servers 326,000 HQ Server 126,000 Aggregates (Impact)  ( asset value x vulnerability) Laptops Desktops Regional Servers HQ Servers Network Infrast.

.033

.30

.05

.90

0 0 1 0 0 0 0 0 0 0 1 0 0 0 1 0 0 1 0 0 0 0 5,500,330 3,500,330 9,329,000 129,000 3000 Software .033

0 0 0 0 0 • Customize matrix to assets & vulnerabilities applicable to case – Compute cost of each asset and put them in the value row – Determine correlation with vulnerability and asset – Compute the sum of product of vulnerability & asset values; add to impact column

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

330

Case Study

Vulnerability/Threat Matrix

• The coefficients of this matrix are usually based on data from the literature, e.g., – if rate of failure of hardware is r f (per unit time) – the number of pieces of hardware is n then – the total number of failed components during a time period is rf*n – the fraction of hardware that fails is r f *n/n= r f • For the current example we will assume data for illustration of the concept – Failure rate of laptops is .001 per day (i.e., one in a thousand laptops encounters hardware failure during a day) – Similarly failure rate of a desktop is .0002 (i.e. 2 in ten thousand desktops would encounter hardware failure in a given day.

– Hardware failure can cause loss of software, however, our assumption is that all software is replaceable from backups

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Vulnerability/Threat Matrix, cont’d.

– We assume that the hardware failure will disrupt the network once every one hundred days – There is 0.3 percent chance that software failure can lead to failure of desktops – We assume that there is a .01 chance of a laptop being stolen, .001 for a desktop, and .0002 for servers.

– There is a very low chance that network equipment is stolen since it is kept in secure rooms (.0001) – When equipment is stolen some software may have been stolen as well – We assume that denial-of-service is primarily targeted at servers and not individual machines – We assume that the denial-of-service can disable machines as well as cause destruction of software – Insider attacks are primarily meant to exploit data & disable machines – We assume that the servers have less access thus are less vulnerable to insider attacks

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Vulnerability/Threat Matrix, cont’d.

Vulnerabilities Threats

Input Impact Aggregates

 Hardware Failure Software Failure Equipment Theft Denial of Service Viruses/Worms Insider Attacks Intrusion Laptops 5,500,330 .001

.003

.0160

.0001

.003

.001

Desktops 3,500,330 .0002

.003

.001

.0001

.003

.001

Regional Servers 9,329,000 .0002

.003

.0002

.001

.003

.0001

.001

HQ Servers 129,000 .0002

.003

.0002

.001

.003

.0001

.001

Network Infrast.

3,000 .01

0 .0001

0 0 .0001

0 Software 330 0 0 .005

0 .001

.001

Aggregates (Threat Importance)  (

impact value x

threat value) 8,122.00

55,375.98

93,399.16

10,358.07

55,376.31

9,947.09

18,458.99

• Complete matrix based on the specific case – Add values from the Impact column of the previous matrix – Determine association between threat and vulnerability – Compute aggregate exposure values by multiplying impact and the associations

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Threat/Control Matrix

• Some of these controls have threats associated with them. However, these are secondary considerations and we will be focusing on primary threats.

• We assume that IDS systems will control 30% of the DOS attacks, 30% of Viruses and Worms and 90% of intrusions – In addition, IDS systems do not impact insider attacks • Anti-Virus Software will prevent 90% of Viruses and Worms.

• That upgrades to a firewall will greatly control (90% each) of DOS attacks, as well as Viruses and Worms. It will control 30% of intrusions, but not insider attacks.

• A redundant HQ server will control 10% of hardware failure (when the original HQ server fails). This is the same percentage for theft and insider attacks.

• Also, a redundant HQ server will help with 80% in cases of DOS attacks on the HQ server.

• Spare laptops will assist in cases of hardware failure and theft (30% because of volume).

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Case Study

Threat/Control Matrix, cont’d.

• We assume that warranties will help with 70% of both hardware failure and software failure. While it will assist with the cost of new hardware or software, will not reduce employee time.

• It is determined that insurance will be able to control 90% of impacts from the threats of theft, DOS attacks, Virus/Worm attacks, Insider Attacks, and Intrusion. • Physical controls (locks, key cards, biometrics, etc.) will control 90% of theft.

• Also, it is assumed that a security policy will assist with 20% of all threats since every policy can have procedures which can assist in prevention.

• Customize matrix based on the specific case – Add values from the threat importance column of the previous matrix – Determine impact of different controls on different threats – Multiply (1-impact) throughout threat column and multiply to threat importance to get values.

University at Albany Proprietary Information

Case Study

Threat/Control Matrix, cont’d.

Threats Controls

Input Threat Importance Values

 Intrusion Detection Anti-Virus Firewall Upgrades Hardware Failure 8,122.00

0 0 0 Software Failure 55,375.98

Theft 93,399.16

Denial of Service 10,358.07

0 0 0 0 0 0 .30

0 .90

Viruses/ Worms 55,376.31

.30

.90

.10

0 .10

.80

0 Redundant HQ Server Spare Laptops Warranties Insurance Physical Controls Security Policy

Calculate Exposure with Controls

 .30

.70

0 0 .20

1,228.05

0 .70

0 0 .20

13,290.24

.30

0 .90

.90

.20

470.73

0 0 .90

0 .20

11.60

0 0 .90

0 .20

31.01

Insider Attacks 9,947.09

0 0 0 .10

Intrusion 18,458.99

.90

0 .30

 (

threat importance

x impact of controls) 36,333.41

49,838.68

64,698.64

0 Aggregates 19,433.28

0 0 .90

0 .20

716.19

0 0 .90

0 .20

103.37

30,456.35

44,448.59

168,785.66

84,059.24

50,207.52

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

•

Case Study

Assignment

Given the matrices and the example case provided, use this same methodology in application to determine the information security risk in your own organization.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Module 4

Cost Benefit Analysis & Regression Testing

• • • •

Cost Benefit Analysis & Regression Testing

Outline

How to use matrices for cost benefit analysis?

How to calculate Risk Leverage?

Applying the case study example Examples – Unauthorized Access – Graphical Cost Benefit Analysis with Regression Testing

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Cost Benefit Analysis

Matrix Cost Benefit Analysis

• The exposure before controls is equal to the summation of the aggregate values for impact value x threat value. (Vulnerability/Threat Matrix) – In this case, the value is equal to: $251,037.60 • The exposure after controls is equal to the sum of all of the multiplied threat importance values. • For example, in the Hardware Failure column, we will take each of the threat importance values and subtract them each from 1. These values should be multiplied together. (Threat/Control Matrix) – This will give us: (1-.10) x (1 - .30) x (1 - .70) x (1 - .20) = 0.15 – This value will be multiplied by the threat importance value: 0.15 x $8,122.00 = $1,218.30 (cost with controls of Hardware Failure) – Do this for all Threat columns and then summate all the values.

– This value is equal to: $15,851.19

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Cost Benefit Analysis

Risk Leverage

• Costs are associated with both: – Potential Risk Impact – Reducing Risk Impact • Risk Leverage is the difference in risk exposure divided by the cost of reducing the risk • Let – r f – r i be the risk exposure after imposing controls be the risk exposure prior to imposing controls – c be the cost of controls Leverage l = (r i -r f )/c • This tells you how many times the reduction in risk exposure is greater then the cost of controls.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Cost Benefit Analysis

Matrix Example

• We are using this equation to calculate cost: – C

= C

+ C

x t – Where C

– C

is the total cost of control i. is the static (one-time) cost of the control.

– C

is the additional cost per day (maintenance, updates, etc.) for the control.

– t is equal to time (if calculating for a year, would equal 365).

• We are assuming cost of control values for this example: – Intrusion Detection: $21,000 x 11 + $160 x 11 x 365 = $873,400 – Anti-Virus: $1,876 x 4,000 (laptops & desktops) + $1,876 x 11 (number of servers) = $7,524,636 + 11 x $160 x 365 = $8,167,036 – Firewall Upgrades: $10,000 x 211 + $160 x 211 = $2,143,760 – Redundant HQ Server: $100,000 + $160 x 365 = $158,400 – Spare Laptops: $2,500 x 200 = $500,000 – Warranties (3 year): $100 x 4,000 (laptops & desktops) + $1000 x 10 (regional servers) + $1,200 (HQ Server) = $411,200 – Insurance: $5,000,000 (per 365 days) – Physical Controls: $5,000 x 211 + $160 x 211 x 365 = $13,377,400 – Security Policy (creation, implementation, enforcement): $640 x 365 = $233,600

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Cost Benefit Analysis

Matrix Example

• Leverage l = (r i -r f )/c – r i = $251,037.60 x 365 = $91,628,724 – r f = $15,851.19 x 365 = $5,785,684.35

– C = $30,864,796 • $251,037 – $15,851.19 / $30,864,796 = .008

• $91,628,724 - $5,785,684.35 / $30,864,796 = 2.78

– The reduction in risk exposure is almost 3x greater than the cost of controls

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Cost Benefit Analysis

Example #4: Unauthorized access

• Scenario: A company uses a common carrier to link to a network for certain computing applications. The company has identified the risks of unauthorized access to data and computing facilities through the network. These risks can be eliminated by replacement of remote network access with the requirement to access the system only from a machine operated on the company premises. The machine is not owned; a new one would have to be acquired.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Cost Benefit Analysis

Example #4: Unauthorized Access

Cost/Benefit Analysis for Replacing Network Access

Item Risk: unauthorized access and use

Access to unauthorized data and programs $100,000 @ 2% likelihood per year Unauthorized use of computing facilities $10,000 @ 40% likelihood per year Expected annual loss (2,000 + 4,000) Effectiveness of network control: 100%

Amount

$2,000 $4,000 $6,000 -$6,000

Cost Benefit Analysis

Example #4: Unauthorized Access Network Control cost:

Hardware (50,000 amortized over 5 years) Software (20,000 amortized over 5 years) Support personnel (each year) Annual cost Expected annual loss (6,000 – 6,000 +54,000) Savings (6,000 – 54,000) +$10,000 +$4,000 +$40,000 $54,000 $54,000 -$48,000

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Regression Testing

Example #5: Graphical Cost Benefit Analysis

• Scenario: This is a case where use of regression testing is being considered after making an upgrade to fix a security flaw. We want to determine if regression testing is economical in this scenario.

• Regression Testing means applying tests to verify that all remaining functions are unaffected by the change.

• Lets refer to the diagram on the following slide, to compare the risk impact of doing regression testing with not doing it. • Upper part of the diagram – the risk of conducting regression testing • Lower part of the diagram – shows the risks of not doing regression testing

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Regression Testing

Example #5: Cost Savings

• In the two cases, one of three things can happen if regression is done:

– We find a critical fault – We miss finding the critical fault – There are no critical faults to be found. • For each possibility – Calculate the probability of an unwanted outcome, P(UO). – Associate a loss with that unwanted outcome, L(UO).

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Regression Testing

Example #5: Calculation

In our example, if we do regression testing and miss a critical fault in the system (a probability of 0.05), the loss could be $30 million. Multiplying the two, we find the risk exposure for that strategy to be $1.5 million. As the calculations in the figure prove, it is much safer to do regression testing than to skip it.

yes Do regression testing?

no P(UO) = 0.75

Find critical fault L(UO) = $0.5M

Risk Exposure $0.375M

P(UO) = 0.05

Don’t find critical fault L(UO) = $30M $1.500M

P(UO) = 0.20

No critical fault L(UO) = $0.5M

$0.100M

P(UO) = 0.05

Find critical fault L(UO) = $0.5M

$0.125M

P(UO) = 0.75

L(UO) = $30M Don’t find critical fault $16.500M

P(UO) = 0.20

No critical fault L(UO) = $0.5M

$0.100M

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

$1.975M

Combined Risk Exposure $16.725M

Cost Benefit Analysis and Regression Testing

Questions 1 and 2

1) What is regression testing?

2) What is the calculated risk exposure for not doing a regression testing, if finding a critical fault has a probability of 0.35 and the loss is estimated at 4.5 million dollars.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

•

Cost Benefit Analysis & Regression Testing

Assignment

Do a cost benefit analysis based on the matrix that you have created for your own organization.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Module 5

Modeling Uncertainties

• •

Modeling Uncertainties

Outline

How do you model?

Monte Carlo Simulation – What is the approach?

– How to model valuation of assets?

– How to model frequency of threats?

– How to model impact of threats?

– How to model controls?

– How to model distribution of risk exposure?

– How to perform a sensitivity analysis for risk exposure?

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Modeling Uncertainties

• Uncertainty exists regarding value that should be assumed by one or more independent variables in the Risk Model. • Contributions to the model’s uncertainty – Lack of knowledge about particular values – Knowledge that some values might always vary • If it cannot be determined with certainty what value one or more input variables in a model will assume, this uncertainty is naturally reflected on the outcome of the dependent variable(s).

• The risk metric is: – not determined by the value of its independent variables (asset values and vulnerabilities, frequency and impact of threats) – a function of the probability distribution of each of these random variables • A good approach to dealing with uncertainty >> simulation

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Monte Carlo Simulation: Approach

• The approach follows the following steps: – Develop risk model – Define the shape and parameters of probability distributions of each input variable – Run Monte Carlo simulation – Build histogram for dependent variables in the model (risk and updated risk) – Compute summary statistics for dependent variables in model – Perform sensitivity analysis to detect variability sources – Analyze potential dependency relationships among variables in model

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Monte Carlo Simulation: Value of Assets

Sample Mean = 50.00

36.88

44.44

52.00

59.55

67.11

Truncated Normal Distribution(mean = 50) • Asset values here are samples and do not represent collected data – In real cases real assets of the organization need to be identified – Value needs to be assigned to the assets

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Monte Carlo Simulation: Frequency of Threats

• Annualized frequency of threats is required to compute the annualized loss expectancy.

• This data can be collected from several sources – Tracking and collecting data from Internal logs – Report from agencies such as CERT

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Monte Carlo Simulation: Impact of Threats

D3 4 Triangular distribution (mode, max=1, min=0) 0.00

0.25

Mean = 0.37

0.50

0.75

1.00

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Monte Carlo Simulation: Controls

H4 0 0.50

Mean = 0. 53 0.75

0.00

0.25

Triangular distribution( mode, max=1, min=0)

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

1.00

Modeling Uncertainties

Monte Carlo Simulation: Risk Exposure Distribution Histogram of Exposure Risk (1000 runs)

30 25 20 15 10 5 0 5610 10627 25677 15643

Risk (in $)

20660 Histogram of Exposure Risk Cumulative Distribution

Cumulative Distribution of Exposure Risk (1000 runs)

1000 900 800 700 600 500 400 300 200 100 0 5610 10627 15643

Risk (in $)

20660 25677

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

45 40 35 30 25 20 15 10 5 0 47

Modeling Uncertainties

Monte Carlo Simulation: Reduced Risk Exposure Histogram of Reduced Exposure Risk (1000 runs)

271 496

Risk (in $)

720 945 Cumulative Distribution Histogram of Reduced Exposure Risk

Cumulative Distribution of Reduced Exposure Risk (1000 runs)

1000 900 800 700 600 500 400 300 200 100 0 47 271 496

Risk (in $)

720 945

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Monte Carlo Simulation: Sensitivity Analysis e q u e n c y A n n u a z e l i d F r

Worms Passw ord Based Attacks Viruses Intrusion Overflow Attacks

Sensitivity Analysis Exposure Risk

-100.0% -80.0% -60.0% -40.0% -20.0% 0.0% 20.0% 40.0% 60.0% 80.0% 100.0%

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Modeling Uncertainties

Questions 1 and 2

1) Why does uncertainty exist within risk analysis?

2) Describe the approach towards Monte Carlo Simulation.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

•

Modeling Uncertainties

Assignment

Using the data provided in the case study, or your own risk analysis, use Monte Carlo Simulation to provide a graphical display.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Appendix

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Analysis

Summary

• Risk Exposure – RISK EXPOSURE = RISK IMPACT x RISK PROBABILITY •Annual Loss Expectancy (ALE) – Identify and determine the value of assets – Determine vulnerabilities – Estimate likelihood of exploitation – Compute ALE – Survey applicable controls and their costs – Perform a cost-benefit analysis

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Quantitative Analysis

Summary Cont’d.

•Risk Aggregation:

* 



j J K

    1

l L

  1  ( 1  (

t q i



ij i I

  1 ) 

)  b

 a



a l

 •Optimization – simple formulation : *   a

j such that

  1 a

 •Cost Benefit Analysis LEVERAGE = (RISK EXPOSURE before reduction

 1 – RISK EXPOSURE

after reduction ________________________________________________ COST OF REDUCTION )

i I

  1

C i

 •Regression Testing –Used for comparing risk impact •Monte Carlo Simulation – 1)Develop risk model, 2) Define the shape and parameters, 3)Run simulation, 4)Build histogram, 5)Compute summary statistics, 6)Perform sensitivity analysis, 7)Analyze potential dependency relationship

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information

Acknowledgements

Grants & Personnel

• Support for this work has been provided through the following grants – NSF 0210379 – FIPSE P116B020477 • Damira Pon, from the Center of Information Forensics and Assurance contributed extensively by reviewing and editing the material • Robert Bangert-Drowns from the School of Education provided extensive review of the material from a pedagogical view.

Sanjay Goel, School of Business/Center for Information Forensics and Assurance University at Albany Proprietary Information