CMPT 982 – Information Privacy
Schedule: M/W/F 9:30 am to 10:20 am
Feel free to write to me at any time: (my full name, no spaces)
xxxxxxx@sfu.ca, and please preface your e-mail with “CMPT 982”.
CMPT 982 – Information Privacy
Grading:
• 10% Participation
• These will be self-assessment quizzes on Canvas
• 30% Paper Reading
• Submit at least 4 paper reviews for the papers listed in the
course website on Canvas
• You can submit up to 6 and the top 4 scores will be used
• 30% Course Project
• Short Paper OR Software+Demo
• 30% Final Exam
• Multiple-choice exam on Canvas
What is privacy?
Who you are What you do
• Privacy is the disassociation between your identity and your behavior
• It is the right to self-expression
What is privacy?
Belief: Privacy is bad because it damages businesses
Problems:
• The collection of user information is not a
compromise of privacy when it is done with
informed consent and the user has control over its
retention
• Requiring businesses to divulge data collection and
usage practices should not be controversial
• Explicit promises on proper data usage help build
up trust between customer and business
“I don’t want privacy. I
want Amazon to learn
about my interests so it
can recommend good
products for me.”
What is privacy?
Belief: Privacy comes at the cost of security
Problems:
• We need to examine whether or not scenarios in
which privacy and security oppose each other are
arbitrarily constructed
• More often than not, the same technological
solutions enhance both privacy and security
• Solutions to compromise privacy for security
usually involve trusting a party that is immune to
attacks – that is inherently insecure
“What if there’s a
terrorist with a timed
bomb and the defuse
code is in his iPhone?
What then?”
What is privacy?
Belief: There is no point in protecting privacy because
we already leak information about ourselves
everywhere
Problems:
• A few small steps can significantly reduce one’s
information leakage
• Improvements in technology and public policy are
being made constantly
• Don’t let perfect be the enemy of good
“Why should I bother
with privacy when my
laptop, phone, IoT, car,
web browser, and WiFi
are all tracking me
anyway?”
Covid-19 Contact Tracing
Privacy & Goals
Goals:
• Warning people if someone they had contact with recently was
infected so they can quarantine and/or be tested
• Saying who was infected is not a goal
Privacy concerns:
• Sharing someone’s health information compromises their privacy
• Sharing someone’s location also compromises their privacy
• It would be nice if we told people where they were infected,
but that is not possible considering this concern
Threat Model
• When you come into contact with someone, they will need to
know that in case you test positive later
• When you are infected, you will have to announce it in a public
way; this message can be read by anyone
• It is tempting to use a trusted third-party – but this is not a
privacy-enhancing technology solution
Cryptographic primitives
A cryptosystem consists of:
• Key(s)
• Encryption mechanism
• Decryption mechanism
Plaintext Ciphertext
Encryption
Keep this:
Idea
• I can announce my presence to nearby devices as ciphertext
• The ciphertext is passed through Bluetooth
• My actual location is never announced
• Only when I am infected will I release the key
Protocol
• Generate a new temporary exposure key (K) every day
• When in range of anyone, send them a rolling proximity identifier
through Bluetooth encrypted as:
• Since we want them to be able to decrypt this later using the same
key, it should be a symmetric encryption (e.g. AES)
• Retain up to 14 temporary exposure keys on your device
EncK(Known string)
https://covid19.apple.com/contacttracing
Protocol
• If testing positive, release all 14 temporary exposure keys to a
public database
• All users regularly check the public database and attempt to
decrypt the RPIs they have stored using those keys
• If a decryption attempt is successful, you will also know how many
days ago you came into contact with that person
• Success = known string observed
Problems?
• Storage – a few hundred million keys is only gigabytes
• Computation – most computation is done on individual devices,
and thousands of AES decryptions per second is no problem
• Information leakage – my RPI doesn’t reveal anything about me,
not does my key
• But what if an eavesdropper listens to RPIs across several locations
to try to track people?
• The fact that the RPI doesn’t change in a day is an undesirable
property
EncK(Known string)
Increasing RPI granularity
• Based on the temporary exposure key K, derive rolling keys RK
every 10 minutes
• This can be done with key derivation functions (a hash)
• Instead of encrypting with K, encrypt with RK
• Still release K when testing positive; recipients compute RK
• Could we have used RK in the first place?
• Yes; but this trick reduces computation/storage by 144 times
EncRK(Known string)
RK = H(K, time)
Overview
Generate K
every day
RK = H(K, time)
every 10 minutes
RPI = EncRK(string)
send constantly to all nearby
Release K
when infected
Store K
up to 14 days
Retrieve all K
constantly
Public database
Receive RPI
store for up to 14 days
RK = H(K, time)
Check if RK matches RPI
attempt to decrypt and observe string
Potential infector
Potential infectee






























































































































































学霸联盟