"We find vulnerabilities in diverse features like one-time passwords, bookmarklets, and shared pass-words. The root-causes of the vulnerabilities are also diverse: ranging from logic and authorization mistakes to misunderstandings about the web security model, in addition to the typical vulnerabilities like CSRF and XSS," they shared in a paper.
The five password managers they analyzed are LastPass, RoboForm, My1Login, PasswordBox and NeedMyPassword, and they did it to evaluate their security in practice, and to provide pointers to "guide the design of current and future password managers."
"Widespread adoption of insecure password managers could make things worse: adding a new, untested single point of failure to the web authentication ecosystem. After all, a vulnerability in a password manager could allow an attacker to steal all passwords for a user in a single swoop," they pointed out, and are advocating a defense-in-depth approach.
Of the five vendors whose products were tested, only the last one (NeedMyPassword) didn't respond when they contacted them and responsibly shared their findings. The other four have fixed the vulnerabilities within days after disclosure.
"Since our analysis was manual, it is possible that other vulnerabilities lie undiscovered," they also pointed out, and announced that they will be working on a tool that automatizes the process of identifying vulnerabilities, as well as on developing a "principled, secure-by-construction password manager."
In the meantime, the LastPass Team has informed its users about the found vulnerabilities with the LastPass bookmarklet and One Time Passwords, the fact that they have fixed them and that they haven't been detected being exploited in the wild, and that there is no danger to users.
"If you are concerned that youíve used bookmarklets before September 2013 on non-trustworthy sites, you may consider changing your master password and generating new passwords, though we donít think it is necessary," they added.
"Regarding the OTP attack, it is a 'targeted attack', requiring an attacker to know the userís username to potentially exploit it, and serve that custom attack per user, activity which we have not seen. Even if this was exploited, the attacker would still not have the key to decrypt user data."