According to the paper (PDF), four out of the five password managers inadvertently leaked a user’s credentials for stored web sites due to all-too-common web based security flaws like Cross Site Request Forgery (CSRF) and Cross Site Scripting (XSS).
The researchers, Zhiwei Li, Warren He, Devdatta Akwawe and Dawn Song, all of the University of California Berkeley, said that they disclosed the holes in August of last year and that all of the affected firms and that all but one – NeedMyPassword – have since patched the vulnerabilities.
Specifically: the researchers found that a remote attacker who could control one or more web servers and DNS domains, and then get a user of the password manager to visit one of those domains could easily hijack a user’s session with the password manager browser plug-in, extracting credentials from managed sites and decrypting them.
Similarly, such an attacker could use a phishing attack to harvest credentials when a user is prompted to log into her master account. Applications like LastPass and Roboform allow their users to authenticate within the active tab, making such an attack feasible.
The researchers said a similar attack could also defeat LastPass’s One Time Password (OTP) login option.
In a blog post response, published on Friday, LastPass said that it didn’t have any evidence that the researchers methods were exploited by anyone else. The company said it has since patched the issues reported.
LastPass customers who used bookmarklets before September 2013 on non-trustworthy sites might want to change their master password and generating new passwords for managed accounts, the company said.
Read more about the research here: The LastPass Blog: A Note from LastPass.