next up previous
Next: Simplified PAK Up: No Title Previous: No Title

Naive Password-Authenticated Diffie-Hellman

First we give a standard Diffie-Hellman key exchange. (For now we won't worry about the specifics of the group in which we are working.)


 
Figure 1: Diffie-Hellman key exchange
\begin{figure}

 {
 \begin{center}

 $\begin{array}
{lcr}
 Alice && Bob \\ 
...
 ... 
 Z = Y^{x} & & Z = X^{y} \\ 
 \end{array}$
 \end{center}
 }

\end{figure}

Of course, this is totally vulnerable to man-in-the-middle attacks. To authenticate the two parties with a password, we could try something like the following,


 
Figure 2: Diffie-Hellman key exchange with authentication
\begin{figure}

 {
 \begin{center}

 $\begin{array}
{lcr}
 Alice && Bob \\ 
...
 ... 
 Z = Y^{x} & & Z = X^{y} \\ 
 \end{array}$
 \end{center}
 }

\end{figure}

However, this is vulnerable to an offline dictionary attack, since both X and k are available to an eavesdropping attacker. So we could try authenticating using the secret value Z instead, as in the following.


 
Figure 3: Another Diffie-Hellman key exchange with authentication
\begin{figure}

 {
 \begin{center}

 $\begin{array}
{lcr}
 Alice && Bob \\ 
...
 ...assword}) \mbox{ then abort}\\ 
 \end{array}$
 \end{center}
 }

\end{figure}

However, while we have secured the protocol against eavesdropping attackers, active attackers can still obtain enough information to perform an offline dictionary attack. (Hint: Imagine an attacker that pretends to be Alice and sends her own X=gx value.) So this is still not sufficient for totally secure password-authenticated key exchange.



Phil MacKenzie
9/29/2000