Privacy in the intelligent ambience and network complexity Miltos Anagnostou School of Electrical and Computer Engineering National Technical University of Athens The objective of this talk • The need for privacy protection is likely to drastically shape future networks and services and to increase their complexity. • Otherwise, the general population will most probably not accept ambient intelligence in every day life. 2 Basic Privacy: The lamp example (I) • Assumptions: One house, one room, one lamp, one human. • What can possibly be the meaning of switching a lamp on and off? (Trivial answer: any digital signal.) • Assume usual meaning: “On” means presence, “off” means absence (or sleep). • What is privacy in this context? An external observer should not be able to detect human presence/absence. • Light can be seen only by nearby observers. 3 The lamp example (II) • Trivial solution: Close window or remove lamp (effective, but degrades “service”). • “Sophisticated” solution: An automatic mechanism is used to produce an on/off pattern in the human’s absence. • Intruder’s attack 1: Analyse on/off pattern, see if it is realistic. • Intruder’s attack 2: Combine information from different sources: “Is car parked outside?” • Countermeasure to attack 2: Have second car (at a cost). 4 The Dinner Example: Bob’s view • Alice and Bob discuss (at 15:00 hrs on the phone) having dinner together. • Bob calls restaurant R (19:00), books table for 2. • Bob picks Alice from home (20:00). • Bob & Alice have dinner together at R (20:30-22:30). 5 The dinner example: Trudy’s view Trudy has access to call data & location update data of Bob & Alice. Trudy observes the following events: • Bob has called Alice (among others) and restaurant R. • At 20:00 Bob and Alice are in the same cell C1. Alice’s home is also in C1. • At 21:30 Bob and Alice are in the same cell C2 (≠C1). Restaurant R (21:30) is in C2. 6 Countermeasure “reduce location updates” • Location updates in GSM are created periodically, on power-on, on call reception. Purpose is to reduce search time upon an incoming call. • Countermeasure: Cancel all periodic updates. • Result: Location changes are not monitored between successive calls. • Disadvantage: Response time for incoming calls is prolonged. • Phone search algorithm is complicated. 7 Enter Ambient Intelligence, alias Pervasive Computing AmI is the vision of an environment • filled with smart and communicating devices, • which are naturally embedded in the environment and in common objects, • while their presence is kept as seamless as possible. 8 Ambient Intelligence applications Smart objects: • Tags, • clothes, • tools. Smart spaces: • Smart floors • Sound based tracking mechanisms • Face recognition devices Smart reasoning, management of smart devices: • The Cooltown model 9 Mediacup project Smart mugs are equipped with • sensors: temperature, motion (acceleration), touchdown (switch under the bottom), content (weight), location • communication and processing capability. Objective: To draw conclusions from using smart objects. 10 Context aware services CA services take into account • Location, space form, lighting, weather and other properties of the environment • time, • present persons, activities, sentiments, • preferences, • Interaction/dependencies with/from other services/applications. Increase convenience while human-machine interaction remains seamless. 11 Smart floor • Tracking mechanism. • Can distinguish between different persons. Can be used for identification. • Can receive signals. 12 The main problem If ambient intelligence is about collecting and communicating information, how can privacy be protected in such a world? 13 Location privacy • Capability to control one’s own location information distribution. • Total location privacy is not desirable: Accidents can happen to an invisible person. • However, modern technology is changing location resolution and intruder population size. • Your location privacy is likely to be compromised each and every time you interact with an intelligent object of known position. 14 Public is not always willing to accept reduced location privacy: The RFID case • Benetton announces intention to use RFID in products (March 2003). • RFID supplier Philips does not publicise after sales deactivation method. • Consumers suspected surveillance capability when purchasing and when using a product. • Consumers boycott Benetton, which retreats. 15 Location v. identity • Some location based services can be used without identification: “When I reach a product I would like to know absolute and comparative price information.” • E-payment services usually require authentication (although there is ongoing research on anonymous money). • A Heizenberg principle would be desirable. 16 Who is Trudy? • A security/privacy study begins with an assessment of possible threats and possible enemies. • In the past only spies and lovers were interested in cryptography. • Today there is not only industrial spying, but also surveillance of the common human’s habits for promotional purposes (e.g. by using cookies). • Law enforcement and national security agencies are increasingly favouring less privacy for more security and safety (safe flights v. safe communications dilemma). 17 The “program” According to a recent (17/8/06) US court decision • “… a secret program … at least by 2002 and continuing today … intercepts without benefit of warrant or other judicial approval, prior or subsequent, the international telephone and internet communications of numerous persons and organizations …” • “Defendants have publicly admitted to the following: (1) the … program exists; (2) it operates without warrants; (3) it targets communications …” • Decision connects loss of privacy with economic impact. 18 The Athens Olympics “wire”-tapping • Publicly announced in Febr. 2006 by the Greek government. • Probably started in August 2004, prior to the Olympic Games. • Continued until March 2005, when uncovered due to a massive SMS rejection alarm. • Voice stream was duplicated and sent to a group of prepaid-card mobile phones (“shadows”) by special software residing in selected exchanges. • Investigation is open. 19 Arguments against privacy: Industry’s view • Companies and organisations are vulnerable to abuse of their services by their clients. • The obvious “solution” is information collection and information exchange. • Example: The program “Teiresias” monitors bank related fraud. It is used in forecasting the behaviour of persons wishing to borrow money. 20 Arguments against privacy: State’s view • Organised crime and terrorism are using more and more advanced technology. • Law enforcement mechanisms should be able to make use of at least equally sophisticated technology. • Life is more valuable and worth preserving than privacy. • Surveillance technologies can save time and money. 21 Increase Complexity @ Network Level • Mix networks (Chaum, 1980) rely on a chain of proxy servers. Each proxy encrypts incoming messages using public key cryptography. • Routing onions (Goldschlag, Reed, and Syverson, 1996) encode routing information in a set of encrypted layers. • Mist routers (J. Al-Muhtadi, R. Campbell et al, 2002): Components, which know user location are separated from components, which know user identity. Disadvantage: No cameras, no talking. 22 Increase Complexity @ Application Level • Crowds is an anonymity agent, which hides a Web surfer's behaviour within a group of other surfers. A request is seen to come from a subset of the total population, which uses Crowds (Reiter, Rubin, 1999). • Lucent Personalized Web Assistant (LPWA): creates and manages pseudonyms. On each return to the same web site the same pseudonym is used. Different pseudonyms are used in different web sites (Gabber, Gibbons et al, 1999). • Anonymous Remailers: Change source & destination IP addresses en route. Create dummy traffic. • Mix Zones: “MZ is a connected spatial region of maximum size in which no user has registered any application callback” (Beresford, Stajano, 2003). 23 Privacy @ application level • Services/applications are not designed for privacy: Skype’s default profile reveals user proximity to PC. • Attack: In an AmI environment Trudy will analyse Bob’s behaviour by combining different channels. • Defense: Devise dummies, which will interact with environment (possibly in different locations). • Counter-attack: Depends on dummy’s intelligence/complexity. • Overhead: Effective user population will be multiplied (to the benefit of network operators and service providers). 24 Pseudonyms • Long term pseudonyms – can be uncovered through surveillance and behaviour analysis. • Short term pseudonyms – require a change management system. – When system response is expected, system must know where to send it. – Cannot be used with preference profiles. Correlation of profiles uncovers pseudonym. – If location technique resolution is high, successive pseudonyms can be correlated. (Change only in mix zones.) – Different pseudonyms of the same user can be correlated by using user status information. – If profile variations are used, small variations create correlation risk, while large variations will create operation problems. 25 Pseudonym related open problems • How should user react in low privacy level environments? • What is a proper privacy measure? • Dummy users: Application layer, network layer and physical (doors must open, purchases must be made) complexity increase. • How is privacy influenced by location resolution and sampling frequency? • Scalability problems? 26 Privacy protection systems • Geopriv: Secure computational objects are used to protect location and other personal data and to enforce a privacy policy. • P3P and Appel: User decision to accept/reject a privacy level or policy is automated. • pawS: Can be used for description and validation of privacy requirements. An incoming user’s agent is given enough data to check whether existing privacy policy is compatible with user preferences. 27 Conclusion • Privacy is a working assumption in today’s economy: Reduce privacy and certain operational costs will increase. • Ambient intelligence will not be accepted if perceived as privacy threat. • Privacy can be defended at network level by increasing complexity. • Privacy can also be protected at application level. Protection at application level is probably more expensive. • Privacy can be subject to “multi-channel” attacks, the only defence being to use expensive dummies. 28