Detecting compromised credentials in a credential stuffing attack
Techniques are provided for detecting compromised credentials in a credential stuffing attack. A set model is trained based on a first set of spilled credentials. The set model does not comprise any credential of the first set of spilled credentials. A first request is received from a client computer with a first candidate credential to login to a server computer. The first candidate credential is tested for membership in the first set of spilled credentials using the set model. In response to determining the first set of spilled credentials includes the first candidate credential using the set model, one or more negative actions is performed.
1. A computer system comprising:
one or more hardware processors;
at least one memory coupled to the one or more hardware processors and storing one or more instructions which, when executed by the one or more hardware processors, cause the one or more hardware processors to:
receive, from a first client computer, a first request that includes a first candidate credential for a first account on a first website;
determine that the first request is part of an attack, and in response, training a first set model based on the first candidate credential, wherein the first set model represents a first set of spilled credentials that comprise the first candidate credential, wherein the first set model does not comprise the first candidate credential;
receive, from a second client computer, a second request that includes a second candidate credential, wherein the second candidate credential matches the first candidate credential;
test for membership of the second candidate credential in the first set of spilled credentials using the first set model;
determine that the second candidate credential is a member of the first set of spilled credentials, and in response:
determine that the second candidate credential is a spilled credential;
perform one or more negative actions.
2. The computer system of claim 1 , wherein the one or more negative actions comprise terminating the second request.
3. The computer system of claim 1 , wherein:
the first request is for a first website computer;
the second request is for a second website computer;
the one or more negative actions comprise forwarding the second request to the second website computer with one or more values indicating that the second request is the attack.
4. The computer system of claim 1 , wherein:
the first request is for a first website computer;
the second request is for a second website computer;
the one or more negative actions comprise forwarding the second request to the second website computer with one or more values indicating that the second candidate credential in the second request is spilled.
5. A method comprising:
receiving, from a first client computer, a first request that includes a first candidate credential for a first account on a first website;
determining that the first request is part of an attack, and in response, training a first set model based on the first candidate credential, wherein the first set model represents a first set of spilled credentials that comprise the first candidate credential, wherein the first set model does not comprise the first candidate credential;
receiving, from a second client computer, a second request that includes a second candidate credential, wherein the second candidate credential matches the first candidate credential;
testing for membership of the second candidate credential in the first set of spilled credentials using the first set model;
determining that the second candidate credential is a member of the first set of spilled credentials, and in response:
determining that the second candidate credential is a spilled credential;
performing one or more negative actions;
wherein the method is performed by one or more computing devices.
6. The method of claim 5 , wherein the one or more negative actions comprise terminating the second request.
7. The method of claim 5 , wherein:
the first request is for a first website computer;
the second request is for a second website computer;
the one or more negative actions comprise forwarding the second request to the second website computer with one or more values indicating that the second request is the attack.
8. The method of claim 5 , wherein:
the first request is for a first website computer;
the second request is for a second website computer;
the one or more negative actions comprise forwarding the second request to the second website computer with one or more values indicating that the second candidate credential in the second request is spilled.