The Emperor's New Password Manager: Security Analysis of Web-based Password Managers

Zhiwei Li, Warren He, Devdatta Akhawe, Dawn SongUSENIX Security 2014 · OpenReview

TL;DR

This paper dives deep into the security of popular web-based password managers, revealing that many are surprisingly vulnerable. The authors found that attackers can steal user credentials from four out of five managers studied, highlighting a critical need for better security practices in this widely adopted technology.

What Problem Does This Paper Address?

In today's digital world, remembering unique, strong passwords for every online service is a significant burden for users. Password managers promise to alleviate this by securely storing and automatically filling credentials. While the concept is appealing, the paper questions the real-world security of these tools, especially web-based ones that operate within the browser. The authors aim to identify common vulnerabilities and provide guidance for building more secure password managers, addressing the gap between the idealized security of password managers and their often-flawed practical implementations.

Methods

The researchers conducted a comprehensive security analysis of five popular web-based password managers. Their methodology involved:

  • Developing a Threat Model: They defined a web attacker who controls web servers and DNS, capable of tricking users into visiting malicious sites. This model was extended to include attackers who could potentially create accounts within the password manager service itself.
  • Identifying Key Vulnerability Categories: They categorized potential weaknesses into four main areas:
  • Bookmarklet Vulnerabilities: Exploiting JavaScript code that runs in the browser via bookmarklets.
  • Classic Web Vulnerabilities: Standard web security issues like Cross-Site Request Forgery (CSRF) and Cross-Site Scripting (XSS).
  • Authorization Vulnerabilities: Flaws in how access control and sharing of credentials are managed.
  • User Interface Vulnerabilities: Weaknesses in how the password manager interacts with the user, particularly concerning phishing.
  • Case Studies: They performed in-depth analyses of specific features and attack vectors within each of the five password managers.
  • Providing Guidance: Based on their findings, they offered recommendations for improving the security of password managers, advocating for a defense-in-depth approach.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Bookmarklets are a Major Weakness: The study found that bookmarklets, often used for cross-platform compatibility, are a significant source of vulnerabilities. Attackers can leverage these to steal credentials by tricking users into running malicious code.
  • Logic Flaws are Pervasive: Beyond typical web vulnerabilities, the researchers uncovered critical logic errors, particularly in credential sharing features. This suggests that even with strong cryptographic foundations, flawed implementation can undermine security.
  • Authentication vs. Authorization Confusion: A recurring theme was password managers confusing authentication (verifying identity) with authorization (determining what an authenticated user can do). This led to vulnerabilities where users could perform actions they shouldn't be allowed to.
  • Phishing Remains a Threat: Despite password managers aiming to mitigate phishing, some were found to be susceptible to attacks that could expose all of a user's credentials, turning a single phishing attack into a catastrophic event.
  • Predictable Identifiers are Dangerous: The use of sequential or easily guessable identifiers for credentials or sharing actions created a pathway for attackers to exploit authorization flaws.

Key Results

The study found critical vulnerabilities in all five password managers analyzed. Specifically:

  • Credential Theft: In four out of the five password managers, attackers could steal a user's credentials for arbitrary websites.
  • Diverse Vulnerabilities: The identified vulnerabilities spanned bookmarklets, classic web attacks (CSRF, XSS), authorization logic, and user interface design.
  • Real-World Impact: The researchers demonstrated how these vulnerabilities could lead to complete account takeover, privacy breaches (e.g., revealing all websites a user has an account on), and denial-of-service attacks.
  • Vendor Response: Four out of five vendors responded quickly to the reported vulnerabilities, and most bugs were fixed within days. One vendor, NeedMyPassword, had not responded at the time of publication.

Limitations and Future Work

The authors acknowledge that their analysis was manual and that other vulnerabilities might exist. They suggest future work could involve:

  • Developing Automated Tools: Creating tools to automatically identify such vulnerabilities.
  • Building Secure-by-Construction Managers: Designing and developing password managers with security as a core principle from the outset.
  • Exploring Browser Mitigations: Investigating the use of browser features like Content Security Policy (CSP) more deeply.
  • Improving Bookmarklet Security: Recommending an iframe-based architecture for bookmarklets to isolate them within the password manager's origin, or for browser vendors to provide more secure bookmarklet APIs.
  • Defense-in-Depth: Advocating for a layered security approach, combining architectural changes, protocol improvements, and browser-level mitigations.

Summary updated

BibTeX
@inproceedings{li2014emperor,
  author = {Zhiwei Li and Warren He and Devdatta Akhawe and Dawn Song},
  title = {{The Emperor's New Password Manager: Security Analysis of Web-based Password Managers}},
  booktitle = {USENIX Security},
  year = {2014},
  url = {https://openreview.net/forum?id=dkdAswWhgX},
}
Figure