CASA: Context-Aware Scalable Authentication

Eiji Hayashi, Sauvik Das, Shahriyar Amini

Jason Hong, Ian Oakley

Human-Computer Interaction InstituteCarnegie Mellon University

Human-Computer Interaction InstituteCarnegie Mellon University

One Fits All?

Devices require the same user authentication regardless of contexts

If Cost Too Much

Stop using authentication system

A Few Could Fit All

How can we choose security lock system for different situations?

Do they provide better security and usability from users’ perspectives?

Context-Aware

Scalable Authentication

•Authenticate users using active factors and passive factors

•Adjust an active factor based on passive factors

•Quantitative way to choose an active factor

Prototype

Outline

• Underlying Model

• Feasibility Analysis (Field Study #1)

• Prototype Evaluation (Field Study #2)

• Security Analysis

• Design Iteration (Field Study #3)

• Conclusion

Outline

• CASA Framework

• Feasibility Analysis (Field Study #1)

• Prototype Evaluation (Field Study #2)

• Security Analysis

• Design Iteration (Field Study #3)

• Conclusion

CASA Framework

Combining Multiple Factors

Combining Multiple Factors

The probability that a person is a legitimate user given a set of signals

Combining Multiple Factors

The probability that a person is NOT a legitimate user given a set of signals

Combining Multiple Factors

Weight that balances false positives and false negatives

Combining Multiple Factors

Authenticate: A user is more likely to be a legitimate user

Combining Multiple Factors

Reject: A user is less likely to be a legitimate user

Naive Bayes Model

Prototype Evaluation(Field Study #2)

Field Study #2

Test system that changes authentication schemes based on location

Choosing an Authentication Scheme

Location Active Factor

Home ?

Workplace PIN

Other Places ?

Naive Bayes Model

Compare Confidence

Type PIN Be at workplace

Type PIN Be at other place

Compare Confidence

Compare Confidence

Compare Confidence

Type PIN Be at workplace

Type Password Be at other place

Compare Confidence

Chosen Authentication Scheme

Location Active Factor

Home ?

Workplace PIN

Other Places Password

Two Conditions

Location w/ PIN w/o PIN

Home PIN None

Workplace PIN None

Other Places Password PIN

Screenshots

Field Study #2

• 32 participants

• 18 to 40 years old (mean=24)

• On their phones

• For 2 weeks

Result: # of Activations

Condition Home Workplace Other Places

w/o PINNone

13.1 (1.4)None

2.5 (0.4)PIN

8.1 (1.1)

w/ PINPIN

24.5 (3.2)PIN

7.1 (1.0)Password15.7 (2.0)

Result: # of Activations

Condition Home Workplace Other Places

w/o PIN 65.8% 34.2%

w/ PIN 66.8% 33.2%

Result: User Feedback

ConditionEasy to

understandSecure Prefer to use

w/o PIN 5 4 3.5

w/ PIN 4 4 3

Quotes

P3 said, “I don't normally use a security lock, but I would be much more inclined to use one if it didn't require constant unlocking.”

Quotes

P5 said, “I like the system. It’s a great pain to type pin at home, because the nature of the phone, it goes to sleep quickly, then I have to type pin again, which is super annoying.”

Quotes

P12 said, “Typing passwords to check text was annoying. I don't think I will use it.”

Appropriate Security Level

Location Using PIN No Security Locks

Home None

Workplace

Other Places PIN

Appropriate Security Level

Location Using PIN No Security Locks

Home PIN

Workplace PIN

Other Places PIN

Appropriate Security Level

Location Using PIN No Security Locks

Home PIN None

Workplace PIN

Other Places PIN

Appropriate Security Level

Location Using PIN No Security Locks

Home PIN None

Workplace PIN None

Other Places PIN None

Design Iteration(Field Study #3)

Design Iteration

• Appropriate security level

• Workplace is not as safe as home

Appropriate Security Level

Location Active Factor

Home None

Workplace

Other Places

Appropriate Security Level

Location Active Factor

Home None

Workplace

Other Places PIN

Workplace is not safe

No Active Factor Be at Home

No Active Factor Be at Workplace

+

+

Workplace is not safe

No Active Factor Be at Home

Type PIN Be at Workplace

+

+

Workplace is not safe

No Active Factor Be at Home

Using Computer Be at Workplace

+

+No Active Factor +

Active Factor Selection

Location Active Factor

Home None

Workplace when using computers None

Workplace when not using computers PIN

Others PIN

Notification

Field Study #3

• 18 participants

• 21 to 40 years old (mean=26.3)

• On their phones and laptops

• For 10 to 14 days

Result: At Workplace

Grey: Computer not usedBlack: Computer used

Result: User Feedback

FeatureEasy to

understandUseful Secure

Prefer to use

Location-based

5 4.5 4 4

Comp-based

4.5 4 3.5 3.5

Notification - 4 - 4

Quote

• P17 said, “It is annoying to use security locks all the time, but whereas if I had such a system which requires pin only at unsecure places its usefulness adds more value when compared to the annoyance caused by it. So, I will definitely use it.”

Conclusion

• Proposed a Naive Bayes framework to combine multiple factors to adjust active authentication schemes

• The framework allowed us to choose active factor in a quantitative way

• Field studies indicated that users preferred the proposed system

Backup

Feasibility Analysis(Field Study #1)

Location as a Signal

• People have their own mobility patterns

• Random people don’t have access to certain places

Field Study #1

• Where do people log in to their phones?

• 32 participants

• 7 to 140 days

PlacePlace Mean Time [%]Mean Time [%] Mean Activation [%]Mean Activation [%]

1 (Home) 38.9 31.9

2 (Workplace) 18.7 28.9

Others 42.4 39.2

Security Analysis

Security Analysis

ConditionKnowledge about target users

Uninformed Informed

Technical expertise

Novice Uninformed Novice Informed Novice

Expert Uninformed Expert Informed Expert

Security Analysis

ConditionKnowledge about target users

Uninformed Informed

Technical expertise

Novice Uninformed Novice Informed Novice

Expert Uninformed Expert Informed Expert

Strangers•CASA is as strong as PIN/password

Security Analysis

ConditionKnowledge about target users

Uninformed Informed

Technical expertise

Novice Uninformed Novice Informed Novice

Expert Uninformed Expert Informed Expert

Family members, Friends, Co-workers•Trusted people•However, users trust co-workers less

Security Analysis

ConditionKnowledge about target users

Uninformed Informed

Technical expertise

Novice Uninformed Novice Informed Novice

Expert Uninformed Expert Informed Expert

Dedicated attackers•Rare, but difficult to prevent•Detection rather than prevention

Adjusting Security Levels

Results: # of Activations

Gray: w/ PINBlack: w/o PIN

Compare Confidence

Result: User Feedback

ConditionEasy to

understandSecure Prefer to use

w/o PIN 5 4 3.5

w/ PIN

4 4 3

3 4

Compare Confidence