Identifying Close Friends on the Internet Randy Baden Bobby - - PowerPoint PPT Presentation

Identifying Close Friends

• n the Internet

Randy Baden Bobby Bhattacharjee Neil Spring from HotNets09

Problem

• Security and privacy are only as good as the

user's ability to identify with whom they wish to communicate

• Consequences of compromised friend edges:

spam, phishing, viruses, privacy leaks, identity theft

• Correctly identifying online social network (OSN)

users is difficult

• Impersonation is easy [Bilge, et al., WWW '09]
• Scale and context is unlike well-studied settings

Impersonation

• Copy public information from a different social

network to create an impostor account

• Clone public information on the same social

network

• Most users will not notice that there are two

identical accounts, even if they have already added the friend

• Optionally, first infiltrate the user's social circle to

invade privacy, then also clone private information

Solution

• Public key infrastructure (PKI)
• Usually: a centralized, trusted certificate authority

publishes public keys for the actors in the system

• No central authority can verify the identity of every

user of an OSN

• Decentralized identity verification
• Users responsible for verifying the identities of their

neighbors in the social graph

• Advantage: OSN users have social knowledge

Exclusive Shared Knowledge

• Ask a question that only the friend can answer
• Some real examples
• Which celebrity do I always get confused with Sean

Penn?

• How much did we spend on drinks last night?

(nearest dollar, no $)

Exclusive Shared Knowledge

• Ask a question that only the friend can answer
• Some real examples
• Which celebrity do I always get confused with Sean

Penn?

– Liam Neeson

• How much did we spend on drinks last night?

(nearest dollar, no $)

Exclusive Shared Knowledge

• Ask a question that only the friend can answer
• Some real examples
• Which celebrity do I always get confused with Sean

Penn?

– Liam Neeson

• How much did we spend on drinks last night?

(nearest dollar, no $)

– 104

Challenges

• Designing a protocol to resist known attacks
• Evaluate security of shared knowledge

Asker Impersonation

When didst we first assemble? 1963 1963 Iron Man incorrectly believes that Loki is Thor Iron Man Thor Loki Consequence: verification is one-way; the asker identifies the askee

Askee Impersonation

What was Henry Pym's wife's name? Uhh... Janet? Janet Impostor guesses the correct answer Iron Man Thor Loki Consequences: answer space should be large; protocol should not reveal information

Protocol

• Desired Properties
• One-way verification: at end, asker learns/confirms

the askee's public key

• Zero-knowledge proof of possession of shared

knowledge

• Interactive: immune to offline dictionary attacks
• Existing protocol: SPEKE [Jablon, Sigcomm96]
• Establishes a secure channel based on a shared

passphrase

• Can be applied over an OSN as a browser extension

Can Users Ask Good Questions?

A user study and a Facebook game

Rules

• Users are rewarded for forming “bonds”
• Users are punished for having their bonds

broken

• Users are rewarded for breaking bonds
• Crowdsourced security penetration testing

Asker Askee Impostor Bond Made +1 per bond +1 per bond

• Bond Broken
• 2 per bond
• 1 per bond

+1 per impostor

Data Collection

• April – June 2009
• 171 registered participants
• 70 did not ask, answer, or try to break bonds
• 92 asked or answered at least once
• 9 only tried to break bonds
• Results consider only the 101 active

participants

Friend Graph

Bond Graph

Ability To Ask

Break Attempts

Friend Stranger All Unsuccessful 50% 44% 94% Successful 5% 1% 6% All 55% 45% 100%

Web of Trust

Web of Trust

Conclusion

• Users can
• Use exclusive shared knowledge to identify one-

hop neighbors in a social network

• Sign and publish identifications to identify multi-hop

neighbors and confirm verifications

– 80% of broken bonds in the experiment also had a good

path

• Enables a social PKI, useful for secure systems
• Bond Breaker is just part of the big picture

Identifying Close Friends

• n the Internet

Randy Baden Bobby Bhattacharjee Neil Spring from HotNets09

Persona

• Distributed and decentralized OSN
• Users choose where to store their data
• Users store data encrypted with ABE

– Key distribution mechanisms to use ABE, public key

crypto, and symmetric key crypto in ways that support OSN communication patterns

– Built on assumption that there's a social PKI

• Users need not trust third parties with data
• Can still provide content-agnostic applications,

which includes most core OSN applications