Phishing detection using uniform resource locators

1. A method for phishing detection using uniform resource locators, the method comprising:

accessing data from one or more of a monitored portion of website data and a monitored portion of emails, the data indicating a suspect Uniform Resource Locator (URL);

assigning a rule score based on partial rule scores of each portion of the suspect URL, the rule score indicating a phishing potential based on URL rules;

determining a uniqueness score of the suspect URL, the uniqueness score indicating a degree of uniqueness of the suspect URL from a plurality of known phishing URLs, wherein said determining the uniqueness score comprises:

using an encoder of an encoder-decoder pair to map the suspect URL to a vector representation;

using a decoder of the encoder-decoder pair to re-create, based on the vector representation, the suspect URL as a recreated URL, wherein one or more of the encoder and the decoder are trained using the plurality of known phishing URLs;

approximating an identity function for the suspect URL to determine a degree of error between the suspect URL and a decoded output of the vector representation from the recreated URL, wherein the degree of error is determined between the suspect URL and the recreated URL, and

assigning a uniqueness score to the suspect URL that is inversely proportional to the degree of error between the suspect URL and the decoded output, and

determining a URL phishing score based, at least in part, on the rule score and the uniqueness score for the suspect URL.

2. The method of claim 1 , further comprising:

determining an entropy score of the suspect URL, wherein the entropy score indicates a probability distribution of characters in the suspect URL, wherein the URL phishing score is further based on the entropy score for the suspect URL.

3. The method of claim 1 , further comprising:

determining a length score of the suspect URL, wherein the length score indicates a relative length of characters in the suspect URL, wherein the URL phishing score is further based on the length score.

4. The method of claim 1 , further comprising:

selecting a next detection engine of a plurality of detection engines for processing the suspect URL, the selecting based at least on the URL phishing score, each of the plurality of detection engines for performing one or more respective investigation actions on the suspect URL to determine a particular issue, wherein said accessing, said assigning the rule scores, said approximating, said assigning the uniqueness score, and said determining are performed by a first detection engine of the plurality of detection engines.

5. The method of claim 1 , wherein said determining the URL phishing score is further based on a cryptographic score that indicates analysis of a cryptographic certificate associated with the suspect URL.

6. The method of claim 1 , wherein the degree of error is lower when the suspect URL is associated with at least one of the plurality of known phishing URLs.

7. The method of claim 1 , wherein said determining the uniqueness score further comprises:

using an encoder of the encoder-decoder pair to map the suspect URL to the vector representation; and

apply the vector representation to a machine learning model to determine anomalous vectors if the vector representation does not match one of the plurality of known phishing URLs;

wherein one or more of the encoder or the machine learning model are trained using the plurality of known phishing URLs.

8. The method of claim 1 , wherein said assigning the rule scores comprises:

assigning a respective partial score to each portion of the suspect URL, the assigning based on a database of known suspect phishing URL portions, wherein said assigning the respective partial scores is based on URL rules that indicate how to score each portion based on context of the suspect URL; and

generating the rule score based on the partial scores.

9. The method of claim 1 , further comprising:

triggering a phishing URL alarm based on the rule score or the uniqueness score, prior to determining the URL phishing score.

10. The method of claim 1 , wherein said determining the URL phishing score comprises:

assigning a weight to each of the rule score and the uniqueness score; and

combining the weighted rule score and the weighted uniqueness score for the URL phishing score.

11. A system comprising:

a non-transitory memory storing instructions; and

a processor configured to execute the instructions to cause the system to:

access data from one or more of a monitored portion of website data and a monitored portion of emails, the data indicating a suspect Uniform Resource Locator (URL);

determine entity specific data for a potential phishing target of the suspect URL;

assign, based on the entity specific data, a rule score for each portion of the suspect URL, the rule score indicating a phishing potential;

determine a uniqueness score of the suspect URL, the uniqueness score indicating a degree of uniqueness of the suspect URL from a plurality of known phishing URLs, wherein determining the degree of uniqueness comprises:

using an encoder of an encoder-decoder pair to map the suspect URL to a vector representation;

using a decoder of the encoder-decoder pair to re-create, based on the vector representation, the suspect URL as a recreated URL, wherein one or more of the encoder and the decoder are trained using the plurality of known phishing URLs;

generating an identity function for the suspect URL;

determining, based on the identity function, a degree of error between the suspect URL and the recreated URL, wherein the degree of error is determined between the suspect URL and the recreated URL, and

determining, based on an inverse proportionality of the degree of error, a uniqueness score of the suspect URL, and

determine a URL phishing score based, at least in part, on the rule score and the uniqueness score for the suspect URL.

12. The system of claim 11 , wherein executing the instructions further causes the system to,

determine an entropy score of the suspect URL, wherein the entropy score indicates a probability distribution of characters in the suspect URL, wherein the URL phishing score is further based on the entropy score for the suspect URL.

13. The system of claim 11 , wherein one or more of the encoder or the decoder are trained using the plurality of known phishing URLs so that the degree of error is lower when the suspect URL is associated with at least one of the plurality of known phishing URLs.

14. The system of claim 11 , wherein executing the instructions further causes the system to,

trigger a phishing URL alarm based on the rule score or the uniqueness score, prior to determining the URL phishing score.

15. The system of claim 11 , wherein said determining the URL phishing score comprises:

assigning a weight to each of the rule score and the uniqueness score; and

combining the weighted rule score and the weighted uniqueness score for the URL phishing score.

16. A non-transitory machine-readable medium having instructions stored thereon, the instructions executable to cause performance of operations comprising:

accessing data from one or more of a monitored portion of website data and a monitored portion of emails, the data indicating a suspect Uniform Resource Locator (URL);

determining entity specific data for a potential phishing target of the suspect URL;

assigning, based on the entity specific data, a rule score for each portion of the suspect URL, the rule score indicating a phishing potential;

determining a uniqueness score of the suspect URL, the uniqueness score indicating a degree of uniqueness of the suspect URL from a plurality of known phishing URLs, wherein said determining the uniqueness score comprises:

using an encoder of an encoder-decoder pair to map the suspect URL to a vector representation;

using a decoder of the encoder-decoder pair to re-create, based on the vector representation, the suspect URL as a recreated URL, wherein one or more of the encoder and the decoder are trained using the plurality of known phishing URLs;

determining an identity function for the suspect URL;

determining, based on the identity function, a degree of error between the suspect URL and the vector representation, wherein the degree of error is determined between the suspect URL and the recreated URL, and

determining a uniqueness score to the suspect URL as an inverse proportionality of the degree of error, and

determining a URL phishing score based, at least in part, on the rule score and the uniqueness score for the suspect URL.

17. The non-transitory machine-readable medium of claim 16 , wherein the operations further comprise:

determining an entropy score of the suspect URL, wherein the entropy score indicates a probability distribution of characters in the suspect URL, wherein the URL phishing score is further based on the entropy score for the suspect URL.

18. The non-transitory machine-readable medium of claim 16 , wherein one or more of the encoder or the decoder are trained using the plurality of known phishing URLs so that the degree of error is lower when the suspect URL is associated with at least one of the plurality of known phishing URLs.

19. The non-transitory machine-readable medium of claim 16 , wherein the operations further comprise:

triggering a phishing URL alarm based on the rule score or the uniqueness score, prior to determining the URL phishing score.

20. The non-transitory machine-readable medium of claim 16 , wherein said determining the URL phishing score comprises:

assigning a weight to each of the rule score and the uniqueness score; and

combining the weighted rule score and the weighted uniqueness score for the URL phishing score.