Training machine learning algorithm(s) to identify leaks of sensitive information
A trained machine learning algorithm receives input data that may contain sensitive information. For example, the input data may be top secret military specifications that are sent as an attachment in an email that is being sent outside of a government computer network. The trained machine learning algorithm is trained with one of: sensitive training data or insensitive training data (or there may be two trained machine learning algorithms where one is trained with the sensitive training data and one is trained with the insensitive training data). The trained machine learning algorithm determines whether the input data contains the sensitive information. In response to determining that the input data contains the sensitive information, an action is taken to prevent release of the input data. For example, the action may be to block the sending of the email.
1 . A system, comprising:
a microprocessor; and
a computer readable medium, coupled with the microprocessor and comprising microprocessor readable and executable instructions that, when executed by the microprocessor, cause the microprocessor to:
receive, by a trained machine learning algorithm, input data,
wherein the input data comprises sensitive information and
wherein the trained machine learning algorithm is trained with at least one of:
sensitive training data; and
insensitive training data;
determine, by the trained machine learning algorithm, that the input data contains the sensitive information; and
in response to determining that the input data contains the sensitive information, take an action to prevent release of the input data,
wherein the input data is to be input into an Artificial Intelligence (AI) algorithm and
wherein the microprocessor readable and executable instructions further cause the microprocessor to:
determine if an address is a known address of the AI algorithm; and
in response to determining whether the address is the known address of the AI algorithm, capturing the input data before the input data is sent to the AI algorithm.
2 . The system of claim 1 , wherein the microprocessor readable and executable instructions further cause the microprocessor to:
identify if at least one source watermark in the input data;
in response to identifying that the at least one source watermark is in the input data, determine an access privilege associated with the at least one source watermark; and
use the determined access privilege to further determine whether to prevent the release of the input data.
3 . The system of claim 1 , wherein if the trained machine learning algorithm is trained with the insensitive training data, the trained machine learning algorithm is further trained with the sensitive training data.
4 . The system of claim 3 , wherein the input data is later determined to not contain the sensitive information and wherein the input data is used to filter out future input data that is the same and/or similar.
5 . The system of claim 1 , wherein if the trained machine learning algorithm is trained with the sensitive training data, the trained machine learning algorithm is further trained with the insensitive training data.
6 . The system of claim 5 , wherein the input data is later determined to not contain the sensitive information and wherein the microprocessor readable and executable instructions further cause the microprocessor to:
add the input data to the insensitive training data; and
retrain the trained machine learning algorithm using the insensitive training data with the added input data.
7 . The system of claim 1 , wherein the trained machine learning algorithm comprises a first trained machine learning algorithm and a second trained machine learning algorithm, wherein the first trained machine learning algorithm is trained with the sensitive training data, wherein the second trained machine learning algorithm is trained with the insensitive training data, and wherein determining that the input data contains the sensitive information is based on an output from the first trained machine learning algorithm and an output from the second trained machine learning algorithm.
8 . The system of claim 7 , wherein the output from the first trained machine learning algorithm is a not match of sensitive information and the output from the second trained machine learning algorithm is a match of insensitive information, and wherein the input data is determined to not contain the sensitive information.
9 . The system of claim 7 , wherein the output from the first trained machine learning algorithm is a match of sensitive information and/or the output from the second trained machine learning algorithm is not a match of insensitive information, and wherein the input data is determined to contain the sensitive information.
10 . The system of claim 9 , wherein the input data is later determined to not contain the sensitive information and wherein one or more of the following occurs: the second trained machine learning algorithm is retrained using the input data and the input data is sent to an input filter.
11 . The system of claim 10 , wherein the second trained machine learning algorithm is retrained using the input data.
12 . The system of claim 10 , wherein the input data is sent to the input filter.
13 . A method, comprising:
receiving, by a trained machine learning algorithm executed by a microprocessor, input data,
wherein the input data comprises sensitive information and
wherein the trained machine learning algorithm is trained with at least one of:
sensitive training data; and
insensitive training data;
determining, by the trained machine learning algorithm executed by the microprocessor, that the input data contains the sensitive information;
in response to determining that the input data contains the sensitive information, taking an action, by the microprocessor, to prevent release of the input data,
wherein the input data is to be input into an Artificial Intelligence (AI) algorithm;
determining, by the microprocessor, if an address is a known address of the AI algorithm; and
in response to determining whether the address is the known address of the AI algorithm, capturing, by the microprocessor, the input data before the input data is sent to the AI algorithm.
14 . The method of claim 13 , wherein if the trained machine learning algorithm is trained with the insensitive training data, the trained machine learning algorithm is further trained with the sensitive training data.
15 . The method of claim 14 , wherein the input data is later determined to not contain the sensitive information, wherein the input data is used to filter out future input data that is the same and/or similar.
16 . The method of claim 13 , wherein if the trained machine learning algorithm is trained with the sensitive training data, the trained machine learning algorithm is further trained with the insensitive training data.
17 . The method of claim 13 , wherein the trained machine learning algorithm comprises a first trained machine learning algorithm and a second trained machine learning algorithm, wherein the first trained machine learning algorithm is trained with the sensitive training data, wherein the second trained machine learning algorithm is trained with the insensitive training data, and wherein determining that the input data contains the sensitive information is based on an output from the first trained machine learning algorithm and an output from the second trained machine learning algorithm.
18 . The method of claim 17 , wherein the output from the first trained machine learning algorithm is a match of sensitive information and/or the output from the second trained machine learning algorithm is not a match of insensitive information, and wherein the input data is determined to contain the sensitive information.
19 . A non-transient computer readable medium having stored thereon instructions that cause a microprocessor to execute a method, the method comprising instructions to:
receive, by a trained machine learning algorithm, input data, wherein the input data comprises sensitive information and wherein the trained machine learning algorithm is trained with at least one of:
sensitive training data; and
insensitive training data;
determine, by the trained machine learning algorithm, whether the input data contains the sensitive information;
in response to determining that the input data contains the sensitive information, take an action to prevent release of the input data,
wherein the input data is to be input into an Artificial Intelligence (AI) algorithm;
determine if an address is a known address of the AI algorithm; and
in response to determining whether the address is the known address of the AI algorithm, capture the input data before the input data is sent to the AI algorithm.
20 . The method of claim 13 , further comprising:
identifying, by the microprocessor, if at least one source watermark in the input data;
in response to identifying that the at least one source watermark is in the input data, determining, by the microprocessor, an access privilege associated with the at least one source watermark; and
using, by the microprocessor, the determined access privilege to further determine whether to prevent the release of the input data.