Safe passwords made easy to use Nicolas K. Blanchard 1 , Leila - - PowerPoint PPT Presentation

safe passwords made easy to use
SMART_READER_LITE
LIVE PREVIEW

Safe passwords made easy to use Nicolas K. Blanchard 1 , Leila - - PowerPoint PPT Presentation

Safe passwords made easy to use Nicolas K. Blanchard 1 , Leila Gabasova 2 , Clment Malaingre 3 , Ted Selker 4 , Eli Sennesh 5 1 IRIF, Universit Paris Diderot 2 Institut de Plantologie et dAstrophysique de Grenoble 3 Teads France 4


slide-1
SLIDE 1

Safe passwords made easy to use

Nicolas K. Blanchard 1, Leila Gabasova 2, Clément Malaingre 3, Ted Selker4, Eli Sennesh5

1IRIF, Université Paris Diderot 2Institut de Planétologie et d’Astrophysique de Grenoble 3Teads France 4University of California, Berkeley 5Northeastern University

Stanford EE Computer Systems Colloquium November 28th, 2018

slide-2
SLIDE 2

Passwords are bad, m’kay ?

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 1/27

slide-3
SLIDE 3

Too many passwords

State of password use:

  • Average user has ∼ 100 accounts
  • Creates 50 passwords per year on average
  • Often counterproductive constraints, avoided by users (e.g. 1@MyPassword)

Because of this:

  • High rate of re-use (75% of users)
  • Lots of sharing (40% of users)
  • Frequent loss of passwords (40% to 60% reinitialised every 3 months)

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 2/27

slide-4
SLIDE 4

Too many passwords

State of password use:

  • Average user has ∼ 100 accounts
  • Creates 50 passwords per year on average
  • Often counterproductive constraints, avoided by users (e.g. 1@MyPassword)

Because of this:

  • High rate of re-use (75% of users)
  • Lots of sharing (40% of users)
  • Frequent loss of passwords (40% to 60% reinitialised every 3 months)

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 2/27

slide-5
SLIDE 5

Authentication methods

Multiple alternatives to secure access:

  • Biometrics: have been durably hackable
  • Defer to a service (Facebook connect): trust issues
  • Physical devices: introduce other vulnerabilities
  • Password managers: single point of failure
  • Passwords re-use: extremely vulnerable

Methods to make passwords better:

  • Salt + variable ending: soon vulnerable
  • Blum’s algorithm: costly
  • Passphrases: not compatible with constraints

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 3/27

slide-6
SLIDE 6

Authentication methods

Multiple alternatives to secure access:

  • Biometrics: have been durably hackable
  • Defer to a service (Facebook connect): trust issues
  • Physical devices: introduce other vulnerabilities
  • Password managers: single point of failure
  • Passwords re-use: extremely vulnerable

Methods to make passwords better:

  • Salt + variable ending: soon vulnerable
  • Blum’s algorithm: costly
  • Passphrases: not compatible with constraints

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 3/27

slide-7
SLIDE 7

Passwords vs Passphrases

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 4/27

slide-8
SLIDE 8

It seems we’re stuck with passwords!

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 4/27

slide-9
SLIDE 9

Constraints

Constraints for a good password management algorithm:

  • High entropy for each password
  • High residual entropy against stolen clear-text passwords
  • Memorable even without frequent use (hence deterministic)
  • Easy to understand by non-Turing-award-winners
  • Compatible with frequent constraints

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 5/27

slide-10
SLIDE 10

Idea: mentally extract entropy from a large secret

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 5/27

slide-11
SLIDE 11

Cue-Pin-Select

High level view :

  • Create one high-enropy passphrase and a 4-digit PIN
  • Create a 4-letter cue for each service
  • Deterministically extract 4 trigrams from the sentence using the PIN and the cue

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 6/27

slide-12
SLIDE 12

Example run

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 7/27

slide-13
SLIDE 13

Main Algorithm

Data: Passphrase P of at least 6 random words PIN K of 4 random digits service name N Result: String S of 12 characters begin From N, create string M of four characters L ← − Length(P), V ← − 0, S ← − “” for i = 0 ; i < 4 ; i + + do X ← − M[i] while X / ∈ P do X ← − letter following X in the alphabet V ← − index of next occurrence of X ∈ P after V V ← − V + K[i] + 3 mod (L) S ← → Concatenate (S, P[V − 2], P[V − 1], P[V]) Print S

Algorithm 1: Cue-Pin-Select

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 8/27

slide-14
SLIDE 14

Security analysis

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 8/27

slide-15
SLIDE 15

Bruteforcing Cue-Pin-Select

Today’s standard for web services : 36-42 bits (30 years at 1000 tries/s). Brute-force against Cue-Pin-Select :

  • Naive against a password → 56 bits
  • Optimised dictionary against a password → 52 bits
  • Naive against passphrase → 210 bits
  • Dictionary against passphrase → 111 bits

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 9/27

slide-16
SLIDE 16

Clear-text attacks

To simplify analysis, Very strong adversary model, who knows:

  • 1+ passwords
  • Length of the passphrase
  • Position of each revealed trigram in the sentence

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 10/27

slide-17
SLIDE 17

Residual entropy (empirical on 10 000 tries)

40 50 60 70 80 Bits of entropy 200 400 600 800 1000 Occurrences

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 11/27

slide-18
SLIDE 18

Residual entropy (empirical on 10 000 tries)

10 20 30 40 50 60 Bits of entropy 100 200 300 400 500 600 Occurrences Two plain-texts Three plain-texts

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 11/27

slide-19
SLIDE 19

User experiment

After 4-day experiment:

  • High initial cost (82s on average), and multiple errors initially
  • Quick speed-up, down to 42s after two days, with pen and paper
  • Increase when shift to mental computation only (86s)
  • Speed-up over the last day (down to 57s), no errors
  • Large variability, 24s-71s

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 12/27

slide-20
SLIDE 20

Adaptability

Algorithm can be extended to handle:

  • Number and special characters
  • Length constraints
  • Frequent changes

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 13/27

slide-21
SLIDE 21

Cue-Pin-Select Summary

Cue-Pin-Select:

  • 52 bits security per password
  • Guaranteed resistance to single clear-text attack, probable resistance to 2-3

clear-text

  • Can create 500+ passwords without high risk of strong partial collision
  • Quick learning process to get under 1 min
  • According to models, strongly memorable
  • Natural extension to handle frequent constraints
  • Other extension to improve security

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 14/27

slide-22
SLIDE 22

How to choose a passphrase ?

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 14/27

slide-23
SLIDE 23

Current methods to make passphrase

First possibility: let people choose them Problems:

  • Sentences from literature (songs/poems)
  • Famous sentences (2.55% of users chose the same sentence in a large experiment)
  • Low entropy sentences with common words

Second possibility: random generation Limits :

  • Small dictionary if we want to make sure people know all words
  • Harder to memorise

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 15/27

slide-24
SLIDE 24

Current methods to make passphrase

First possibility: let people choose them Problems:

  • Sentences from literature (songs/poems)
  • Famous sentences (2.55% of users chose the same sentence in a large experiment)
  • Low entropy sentences with common words

Second possibility: random generation Limits :

  • Small dictionary if we want to make sure people know all words
  • Harder to memorise

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 15/27

slide-25
SLIDE 25

What if we take the best of both world ?

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 15/27

slide-26
SLIDE 26

Passphrase choice experiment

We show 20 or 100 words to users, they have to pick – and remember – six. Questions :

  • What factors influence their choices ?
  • What is the effect on entropy ?
  • What are the most frequent mistakes ?
  • How is memorisation affected ?

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 16/27

slide-27
SLIDE 27

Protocol

Simple protocol :

  • Show a list of 20/100 random words from a large dictionary
  • Ask to choose and write down 6 words (imposed on the control group)
  • Show them the sentence and ask them to memorise, with little exercise to help

them.

  • Distractor task: show them someone else’s word list and ask to guess the word

choice

  • Ask them to write the initial sentence

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 17/27

slide-28
SLIDE 28

Interface

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 18/27

slide-29
SLIDE 29

Choosing models

Three main models to analyse user’s choice Uniform : every word with equal probability Smallest : Take the six most frequent words from the list shown Corpus : every word taken with probability proportional to its use in natural language. The word of rank rk is taken with probability :

1 rk

∑n

i=1 1 ri Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 19/27

slide-30
SLIDE 30

Error comparison

Section Correct Typo Variant Order Miss Wrong 1:20 19/47 6 8 6 26 5 1:100 26/51 10 5 3 16 4 Control 6/26 11 11 10 31 12 2:20 14/29 1 2 8 3 2:100 15/26 4 2 3 1 4

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 20/27

slide-31
SLIDE 31

Semantic bias

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Relative word rank in the array 0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14 Proportion of words chosen for each rank 20 words English group 20 words foreign group Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 21/27

slide-32
SLIDE 32

Semantic bias

1-5 6-10 11-15 16-20 21-25 26-30 31-35 36-40 41-45 46-50 51-55 56-60 61-65 66-70 71-75 76-80 81-85 86-90 91-95 96-100 Relative word rank in the array (20 buckets of 5) 0.00 0.05 0.10 0.15 0.20 0.25 Proportion of words chosen for each rank 100 words English group 100 words foreign group Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 21/27

slide-33
SLIDE 33

Syntactic bias

Syntactic effects :

  • Average frequency (< 50%) of meaningful sentences
  • 65 different syntactic structures for 99 sentences
  • Single frequent structure: six nouns in a row

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 22/27

slide-34
SLIDE 34

Syntactic bias

1 2 3 4 5 6 Word position in the passphrase 0.0 0.2 0.4 0.6 0.8 1.0 Proportion of each grammatical category noun adjective verb verb (past tense) gerund adverb Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 22/27

slide-35
SLIDE 35

Entropy comparison

Strategy Entropy (bits) Strategy Entropy Uniform(87,691) 16.42 Smallest(20) 12.55 Corpus(13) 16.25 Uniform(5,000) 12.29 Corpus(17) 16.15 Uniform(2,000) 10.97 Corpus(20) 16.10 Smallest(100) 10.69 Corpus(30) 15.92 Corpus(300,000) 8.94 Corpus(100) 15.32 Corpus(87,691) 8.20 Uniform(10,000) 13.29

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 23/27

slide-36
SLIDE 36

Entropy curves

20000 40000 60000 80000 Rank of n in the dictionary (sorted by decreasing frequency) 0.0 0.2 0.4 0.6 0.8 1.0 P(X ≤ n) Smallest(20) Corpus(100) Corpus(30) Corpus(20) Corpus(17) Corpus(13) Group 20 Group 100

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 24/27

slide-37
SLIDE 37

Conclusion

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 24/27

slide-38
SLIDE 38

Passphrase choice method

Advantage with 100-word list:

  • Secure: 97% of maximal entropy, 30% increase over uniform with limited dictionary
  • Memorable: error rate divided by 4
  • Lightweight: <1MB tool, can and should be used inside a browser

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 25/27

slide-39
SLIDE 39

Questions

Questions on passphrase choice:

  • What is the optimal number of words to show ?
  • Is it interesting to take even bigger dictionaries ?
  • Can this method and Cue-Pin-Select be applied to languages with small

vocabularies (Esperanto)

  • What is the best way to model user choice ?

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 26/27

slide-40
SLIDE 40

Future password research

Our recent work:

  • Typo-tolerant password checkers
  • Culturally neutral codes and passwords for e-voting

Planned research:

  • Better attack models with fewer assumption to prove higher resistance
  • Mental computing cost model to test password algorithms without user

experiments

  • Alternative to Cue-Pin-Select that works in <30s

Introduction Cue-Pin-Select Security and usability Passphrase choice Empirical results Entropy Conclusion 27/27