A large-scale study of password-protected Web sites revealed a lack of standards across the industry that harms end-user security, according to two researchers working at the University of Cambridge in England.
In particular, the weak implementations of password-based authentication at lower-security sites compromises the protections offered at higher-security sites because individuals often re-use passwords, Joseph Bonneau and Soren Preibusch asserted in a paper presented at the Workshop on the Economics of Information Security in Cambridge, Massachusetts, Monday.
Attackers can use low-security Web sites such as news outlets to figure out passwords associated with certain e-mail addresses, and then use those passwords to access accounts at higher-security sites such as e-commerce vendors, Bonneau said.
In an effort that the researchers said is the largest empirical investigation into password implementations to date, they collected data from 150 Websites and found widespread “questionable design choices, inconsistencies, and indisputable mistakes,” according to Bonneau and Preibusch.
The researchers seemed disinclined to blame users for re-using passwords or making them easy to guess, arguing that most users have too many online accounts to manage them all securely.
“Sites’ decisions to collect passwords can be viewed as a tragedy of the commons, with competing Web sites collectively depleting users’ capacity to remember secure passwords,” they wrote.
The large majority—78 percent—of sites examined failed to provide users with feedback or advice on choosing a strong password. Only five sites let the user register password hints, a strategy that will encourage users to come up with stronger passwords. Just seven sites required users to mix numbers and letters, and only two demanded that passwords include non-alphanumeric characters as well.
Bonneau and Preibusch also identified widespread weaknesses in how passwords are submitted to the server when users log in. Only three sites used techniques that prevent the server from receiving a user’s cleartext password at login, although two of those collected cleartext passwords at enrollment.
Most of the sites, 126 in all, seemed to allow unlimited attempts to guess a password; the researchers used a script that attempted incorrect guesses 100 times, after which a person typing in the correct password was able to log in successfully. This indicates that most sites don’t bother to protect against guessing attacks, they said.
Overall, recognized best practices in password security are widely ignored, Bonneau and Preibusch said. Well over half the sites failed to use TLS (transport layer security) to protect password transmission at every stage—some used it at enrollment but not at the login or update point, for example.
In what Bonneau called the “worst practice in the industry”, 29 percent of sites tested e-mailed users cleartext passwords. In addition, 83 percent allowed unrestricted probing for user membership, and 84 percent permitted unrestricted password guessing.
The sites whose owners are the worst offenders are content sites as newspaper Web sites which don’t tend to store sensitive user information, the researchers said. Conversely, sites that store payment details had significantly stronger security practices, the researchers said.
So why do so many sites collect passwords when the practice is generally harmful when poorly implemented? Those sites with poor password practices also seem to be those with an interest in collecting e-mail and personal data about their users.
While broader adoption of delegated protocols such as OpenID would help, Bonneau and Preibusch are pessimistic that the market will support such solutions at the cost of these opportunities to collect user information.