Web Browser Privacy and Security Part I Today’s Topics Trusted Paths Context-Sensitive Certificate Verification (optional paper) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Trusted Paths Trusted paths are used to help users ensure that they are communicating with whom they think they are • Ex. Ctrl-Alt-Del in Windows systems cannot be intercepted Trusted paths for Web are difficult because • From remote server to browser to user • Trivial to make fake UIs that look legit Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Example Attack #1 Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Example Attack #1 Is this from eBay? No trusted path, hard to tell Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Example Attack #2 Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Example Attack #2 Is this from eBay? No trusted path, hard to tell Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Example Attack #3 Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Example Attack #3 Is this from eBay? No trusted path to real eBay to verify Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ One Idea: Dynamic Security Skins User remembers one image • Shown in a trusted window User remembers one password • Ease of use • Sites get hashed password only Uses Secure Remote Password w/ server • Generated using a shared secret Dhamija and Tygar, The Battle Against Phishing: Dynamic Security Skins, SOUPS 2005 Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ How to Show Trusted Path Static security indicators • Ex. Secure window uses a certain color border • Ex. Secure window uses lock icon • Rejected, too predictable and easy to spoof Custom security indicator • Ex. One indicator per site • Ex. One indicator per user • Rejected, too much effort • (Also too much to remember) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Dynamic Security Skins • In theory, lots of images should make it hard to spoof • Trusted path to password window Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Dynamic Security Skins • A unique pattern is generated by each web site (visual hash) • Trusted path from password entry to web site Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Another Idea: Tokens Two factor authentication • Something you have • Usually cryptographic SecureID Smart cards Random cryptographic tokens Scratch cards Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ A Third Idea: Mobile Phones Everyone’s got a mobile phone Client side certificates • Private keys generated/stored on phone • New key for each phone Keys linked to domain names Key generated upon new connection Bluetooth from phone to PC Very few server modifications Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Discussion of Trusted Path “[O]n each launch of Firefox, paint the Firefox interface with a nonintrusive, randomly generated pattern. Because sites wouldn’t be able to replicate this pattern, users would know when they were viewing [a] spoofed UI” Other ideas for trusted paths? Other barriers to adoption? Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Today’s Topics Trusted Paths Context-Sensitive Certificate Verification (optional paper) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Certificates A secure way of binding a public key with an identity • Ex. Amazon sends its certificate via https • Makes it easier to encrypt communications How to know if this certificate is legitimate? • Certificate is also signed by a well-known certificate authority (CA) • Certificates of these CAs often included in web browser Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Self-Signed Certificates Some sites use self-signed certificates • Want to avoid monetary and overhead costs • Often leads to security alerts like below Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Why Certificate Verification Fails 1. Browser may not know public key of the CA that issued the server’s certificate • Internal web server (only by members of the organization) (significant annual fee) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Why Certificate Verification Fails 1. Browser may not know public key of the CA that issued the server’s certificate • Internal web server (only by members of the organization) (significant annual fee) • Own CA: public key installed in browser (no verification errors), but large number of users / user owned computers means high maint 2. Issuer’s or the server’s certificate may be expired Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Why Certificate Verification Fails 3. Common name of certificate does not match server’s fully qualified domain name ♦ Mistake, ex. s3.acme.com vs s10.acme.com ♦ Might be attacker using his own identity with a CA generated certificate (difficult) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Aside: Phishing Attack Signed certificate from Equifax / Geotrust Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Why Certificate Verification Fails 3. Common name of certificate does not match server’s fully qualified domain name ♦ Mistake, ex. s3.acme.com vs s10.acme.com ♦ Might be attacker using his own identity with a CA generated certificate (easy, but expensive) ♦ Might be attacker using a stolen certificate (along with the private key) (difficult) ♦ Or might be self-signed certificate (easy) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Why Certificate Verification Fails Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Why Certificate Verification Fails Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Discussion Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Context-Sensitive Certificate Verification Clarify relationship between user and server’s (non verified) certificate • Not giving the user override mechanisms Distribute signed certificates of internal servers out of band Use typically unused certificate fields: • CA’s contact information (field: issuer alternative name) • CA administrator’s name, address, telephone and fax numbers, and work hours. Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Context Sensitive Certificate Verification Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ If you said you are an internal member… Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ If you said you are an external member… Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Specific Passwords Warnings Helps prevent eavesdropping Allow overriding Existing version: Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Specific Passwords Warnings Is this an important account? Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Specific Passwords Warnings Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Discussion Thoughts so far on designs? • Context-sensitive Certificate Verification • Specific Password Warnings Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ User Studies Computer literate users (CLU) Evaluate: • Likelihood of successful attack in representative security-sensitive Web apps • Possibility of “foolproofing” browsers, so they can be used securely even by untrained CLUs • Can education about the relevant security principles, attacks, and tools improve the security of how users browse the Web? Note: This last hypothesis is not covered in this presentation Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Study’s Design 17 male participants (Pitt CS seniors) Two studies: • Unmodified browser (IE) • Modified Mozilla Firebird 0.6.1 with CSCV and SPW No feedback given between these two studies • (Note: ordering not randomized) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Study’s Design Visit three fictional but realistic sites • Students given password protected accounts Site1: “maintained by Pitt” • Monitor reward points (do well in exams, etc) • HTTPS + Certificate issued by internal CA Site2: “e-merchant not affiliated with Pitt” • Spend reward points on books, CDs, etc. • HTTPS + bogus certificate Site3: “users’ Web email accounts” • HTTP only (no certificate) Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Study’s Design User’s Action Access to a site despite lack of security Simply did not visit the site insecurely Correctly obtained and installed the issuing CA’s certificate Choosing not to access to 2nd and 3rd site insecurely Score (points) 0 50 100 100 Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Study’s Results Guesses? Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Study’s Results With current Web browsers, the mentioned attacks are alarmingly likely to succeed • More often than not, users’ behavior defeats the existing Web security mechanisms. • “um, another of those pop-ups.” • “I always just click yes when I see these popups.” Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Study’s Results CSCV blocked MITM attacks against HTTPS-based applications completely SPW greatly reduced the insecure transmission of passwords in an HTTPbased application Although untrained, users had little trouble using CSCV and SPW Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Discussion Thoughts on results? Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/ Discussion Possible novelty effects • People might change behavior after getting used to new messages Behavior outside of lab study • People might still not go find person to verify Usable Privacy and Security • Carnegie Mellon University • Spring 2007 • Cranor/Hong • http://cups.cs.cmu.edu/courses/ups-sp06/