1. Multiple Password Interference in text Passwords and click based Graphical Passwords by Sonia Chiasson, Alian Forget, Elizabeth Stobert, PC van Oorschot and Robert Biddle Presented by: Payas Gupta

2. Motivation We know that people generally have difficulty remembering multiple passwords. To compare multiple text password recalls with recall of multiple click- based graphical password. – Short term – Long term

3. What it is about? No algorithm no technique It has only user study. But a message as how to show such results in a nice way

4. PassPoints 5 click points in the same order Tolerance accepted around each click point

5. Hotspots Dictionary attacks in graphical password: – Areas of the image that have higher probability of being selected by users.

6. Study Details Hypothesis – Click based graphical passwords would be easier for users to recall than text passwords when users had multiple passwords to remember. – Less interference from multiple unique graphical passwords than multiple unique text passwords.

7. Specific hypothesis Participants will have lower recall success rates with text passwords than with PassPoints passwords. Participants in the Text condition are more likely than PassPoints participants to use patterns across their own passwords. Participants will recall text passwords more slowly than PassPoints passwords. Participants in the Text condition are more likely than Pass-Points participants to create passwords that are directly related to their corresponding accounts. Participants in the Text condition will make more recall errors than participants in the PassPoints condition.

8. Demographics 65 participants – 26 males and 39 females Participants were primarily university students from various degree programs. None were expert in computer security

9. Methodology 65 participants in session 1 Second session after two weeks – 26 participants

10. Session 1 Create Confirm Answer Questions – Perceived difficulty of creating Perform Distraction Task – Mental rotation test Login – Retry as many times to get it correct

11. Results Used chi-square test to compare non- ordered categorical data (comparing login/failure ratios). Success rate – The success rate is the number of successful password entry attempts divided by the total number of attempts, across all participants.

12. Recall 1 First attempt – Text passwords – 68% – PassPoints – 95% Participants could try recalling their password as many times as they wished, until they either succeeded or gave up. Participants in the Text condition reached an 88% success rate with multiple recall attempts, compared to 99% for PassPoints participants.

13. Recall 2 Two weeks after creating their passwords, only 70% of Text participants and 57% of PassPoints participants were able to successfully recall their passwords. Higher accuracies in male in passpoints. – Result aligns with psychology research – Male tend to perform better in visual and female in linguistic tasks

14. Recall Errors

15. Success rate for male and female Recall 2

17. Timings Recall-1 – Participants were quicker at entering PassPoints passwords and this aligns with the fact that participants made fewer errors in the passpoints condition (when participants repeatedly entered the passwords). Recall-2 – No significant difference

18. Use of Mnemonics 23 out of 34 (68%) participants in the Text condition used the account as a cue for at least one of their passwords. – Some passwords were directly linked with the account name. – instantmsg for the instant messenger – “lovelove” for the online dating account – 40% of text passwords were related to their account – males being more likely to create passwords that were directly related to their accounts

19. For text conditions Recall 1 – Participants classified as having used account-related text passwords had a 96% success rate for Recall-1 while those who did not had an 83% recall success rate. Recall 2 – Those classified as having created account-related passwords had a 71% success rate for Recall-2, while those who did not had a 69% success rate.

20. Text Password Patterns 71 out of 204 passwords (35%) were obviously related to other passwords created by the same user – ins901333” for the instant messenger account and “lib901333” for the library account

21. PassPoints Patterns The earlier study found that in PassPoints, participants were likely to select click-points in simple patterns such as a straight line or C- shape

22. Comparison PPLab and MPP

23. Found no statistical difference between the patterns found in the current study (where participants had to create and remember multiple passwords) and the earlier PassPoints lab study (where participants had to remember only one password at a time). Two participants had 4 out of 6 passwords following a “Z” pattern

24. Text Password Dictionary Attack First tested passwords using the free dictionary of 4 million entries. Followed by a second attack using a larger dictionary of 40 million entries purchased from the John the Ripper web site. Smaller cracked 9.8% Larger cracked 15.2%

25. Examples of passwords that were not cracked by John the Ripper include: “msnhotmail” for an email password, “instantmsg” for an instant messenger account, and “inlibrary” for a library account. In an earlier study of text passwords [16], 9.5% (18 out of 190) of passwords were cracked using John the Ripper with the same 4 million entry dictionary and 18.9% (36 out of 190) of passwords with the larger dictionary.

26. Passpoints hotspot formation To evaluate PassPoints passwords for predictability, we compared the distribution of click-points in the current study to those of an earlier PassPoints study on the same images [6]. – Wanted to see whether there was increased clustering of click-points across participants.

27. The J-function measures the level of clustering of points within a dataset. – 32 PassPoints participants for each image in this study (160 click-points per image). – The earlier PassPoints datasets [6] contained between 155 to 220 click-points per image.

28. J-stat

29. Validation of hypothesis Participants will have lower recall success rates with text passwords than with PassPoints passwords. – Hypothesis partially supported. Participants in the Text condition are more likely than PassPoints participants to use patterns across their passwords. – Hypothesis partially supported. Participants will recall text passwords more slowly than PassPoints passwords. – Hypothesis partially supported.

30. Participants in the Text condition are more likely than PassPoints participants to create passwords that are directly related to their corresponding accounts. – Hypothesis supported. Participants in the Text condition will make more recall errors than participants in the PassPoints condition. – Hypothesis supported.

31. Not a mirror image of real life Unlikely to create 6 passwords one at a time No one in our study wrote down their password, users often tend to do so. However, examining the issue of multiple password interference in a controlled laboratory setting is an important step in understanding the effects of increased memory load and the coping behaviours exhibited by users.