At Byron Bush Bash having Beers with Bitcoiners #bbb

stack sats

Can you hear me now?

Attacks on Anonymity Systems: The Practice Roger Dingledine Len Sassaman Attack Methodology • What attacks really exist? • When building anonymity systems: – Designers anticipate possible attacks, and try to protect against them – Many of these attacks may not be feasible – Some may not be preventable • Implementers must focus first on thwarting attacks that are most likely to be used Threat models • In order to evaluate an anonymity system, one must know the threats it addresses • What are the attacker(s) capabilities? • What kind of damage is acceptable? • What are the reasonable performance, reliability, and price trade-offs? Types of Adversaries • Global Observer – Has omniscient network view (and can process data effectively!) • External Attacker – No special advantages – Can send messages into system, observe output • Rogue Operator – Owns a node and knows the business Attack Goals • Break anonymity – Compromise selective users – Compromise all users – Conditional anonymity • Break utility – Prevent anonymity service from being reliable – Redirect potential users to less secure services – Breaking utility leads to breaking anonymity – Deny service entirely Anonymity Breaking Attacks • Replay Attacks • Blending Attacks • Attacks on multiple messages / large files • Pseudospoofing • Tagging attacks • Partitioning attacks (passive and active) • Intersection attacks • Timing and packet counting attacks Replay Attacks • “Déjà vu” • A captured message will follow the same path when resent • Traceable by a Global Observer • Provides clues to Rogue Operators, and possibly to External Observers – Posts to Usenet, etc. Blending Attacks • “an unfriendly crowd” • Trickle, flood, n -1 • Intended to defeat a mix • Requires observation capabilities • Requires traffic flow manipulation Attacks on Multiple Messages • “We’re not like everyone else” • Large files become multiple messages • Traffic analysis is easier • Input and exit correlation • Mix network can be a black box Pseudospoofing • “tentacles and sock-puppets” • An attacker running many nodes increases the chance of chains consisting of entirely his nodes • Users don’t know operators are all one entity acting as multiple personas Tagging attacks • “shuffling a marked deck” • Bit flipping • Tracking identifying markings • Allows blind-spots in observable network Partitioning attacks • “divide and conquer” • Key rotation • Node list discrepancies • Capability changes • Uniquely identifiable clients – (compatibility isn’t the issue -- anonymity system components must operate identically) Intersection attacks • “it’s only a matter of time” • Usage pattern data over time Timing and Packet Counting Attacks • Statistical analysis of network traffic • Low-latency systems at great risk Utility Breaking Attacks • Economic/incentives attacks • Reputation attacks • Flooding attacks Economic/Incentives Attacks • Drive users to less secure systems • Increase cost of more secure systems • Discourage committed operators • Less users mean less security Reputation Attacks • “a good old smear campaign” • Cast doubts on security of strong systems • Spread FUD to less informed users • Discourage development of software and operation of services by targeting principal contributors • Cause confusion Flooding Attacks • Exhaust node resources • Harm node reliability • Create abuse complaints • Can be economic or reputation attacks • Often exacerbated by protocol mistakes – Ex.: Cypherpunk Remailers Capabilities of Attackers • Three types of attackers – Global Observer – External Attacker – Rogue Operator • Three sets of goals – Compromise of Anonymity – Denial of Service – Degradation of Utility Determining Threat Model • Attackers will pick the type of attack which most easily achieves their goals • Anonymity systems should identify the user’s needs as well as his potential adversaries Building the Perfect Anonymity System • Systems which sacrifice usability and reliability in order to protect against attacks that are not able or likely to be used are flawed • Systems should strive for the strongest threat model possible within the existing constraints – “zero-cost improvements” • Remember: More users means more anonymity