Security threat mitigation

1 . A computer-implemented method, comprising:

receiving a system description input pertaining to a target system;

generating a root security query using a machine learning (ML) generator, based on the system description input;

extracting from a security database a security requirement associated with the root security query;

outputting via a user interface an indication of the root security query;

receiving first user input responsive to the indication of the root security query;

generating using the ML generator, based on the first user input and the security requirement, a follow-up request;

outputting the follow-up request via the user interface;

receiving second user input responsive to the follow-up request; and

determining using the ML generator, based on the second user input, that the security requirement is not satisfied by the target system.

2 . The computer-implemented method of claim 1 , comprising performing a security mitigation action responsive to determining that the security requirement is not satisfied.

3 . The computer-implemented method of claim 2 , wherein the security mitigation action comprises generating an alert at a user interface, modifying a setting or parameter of the target system, or isolating a component of the target system.

4 . The computer-implemented method of claim 1 , comprising:

parsing the system description input, resulting in a parsed output;

programmatically generating a query prompt to the ML generator based on the parsed output, resulting in the root security query.

5 . The computer-implemented method of claim 4 , comprising programmatically generating:

a first query prompt based on the parsed output, and

a second query prompt based on the parsed output and a predetermined security topic.

6 . The computer-implemented method of claim 5 , wherein:

programmatically generating the first query prompt comprises programmatically iterating over a set of extracted facts of the parsed output; and

programmatically generating the second query prompt comprises programmatically iterating over a predetermined set of security topics.

7 . The computer-implemented method of claim 1 , wherein generating the root security query comprises:

generating a first candidate root security query and a second candidate root security query using the ML generator, based on the system description input;

generating a first vector embedding for the first candidate root security query and a second vector embedding for the second root candidate security query;

calculating a similarity between the first vector embedding and the second vector embedding;

determining that the similarity between the first and second vector embeddings exceeds a threshold; and

responsive to determining that the similarity exceeds the threshold selecting the first candidate root security query as the root security query.

8 . The computer implemented method of claim 7 , wherein generating the first candidate root security query and the second candidate root security query using the ML generator based on the system description input comprises:

parsing the system description input, resulting in a parsed output;

programmatically generating one or more query prompt to the ML generator based on the parsed output; and

applying the ML generator to the one or more query prompt.

9 . The computer-implemented method of claim 1 , wherein extracting the security requirement associated with the root security query comprises generating a requirement prompt, to query an indexed set of security requirements stored in a long-term memory of the ML generator.

10 . The computer-implemented method of claim 1 , wherein the ML generator comprises one of: a generative pretrained transformer (GPT) model, BLOOM, or LLAMA.

11 . The computer-implemented method of claim 1 , wherein receiving the system description input comprises receiving an image representation of the target system.

12 . A computer system comprising:

a memory configured to store computer-readable instructions; and

a hardware processor coupled to the at least one memory, wherein the computer-readable instructions are configured to cause the hardware processor to:

receive a system description input pertaining to a target system;

generate a root security query using a machine learning (ML) generator, based on the system description input;

extract from a security database a security requirement associated with the root security query;

initiate with the ML generator a conversation session associated with the root security query;

output in the conversation session an indication of the root security query;

receive in the conversation session first user input responsive to the indication of the root security query;

based on the first user input and the security requirement, output in the conversation session a follow-up request;

receive in the conversation session second user input responsive to the follow-up request;

based on the second user input, determine that the security requirement is not satisfied by the target system.

13 . The computer system of claim 12 , wherein the computer-readable instructions are configured to cause the hardware processor to perform a security mitigation action responsive to determining that the security requirement is not satisfied.

14 . The computer system of claim 13 , wherein the security mitigation action comprises generating an alert at a user interface, modifying a setting or parameter of the target system, or isolating a component of the target system.

15 . The computer system of claim 12 , wherein the computer-readable instructions are configured to cause the hardware processor to:

parse the system description input, resulting in a parsed output; and

programmatically generate a query prompt to the ML generator based on the parsed output, resulting in the root security query.

16 . The computer system of claim 15 , wherein the computer-readable instructions are configured to cause the hardware processor to programmatically generate:

a first query prompt based on the parsed output, and

a second query prompt based on the parsed output and a predetermined security topic.

17 . The computer system of claim 12 , wherein the computer-readable instructions configured to cause the hardware processor to generate the root security query by:

generating a first candidate root security query and a second candidate root security query using the ML generator, based on the system description input;

generating a first vector embedding based on the first candidate root security query;

generating a second vector embedding based on the second candidate root security query;

calculating a similarity between the first vector embedding and the second vector embedding;

determining that the similarity between the first and second vector embeddings exceeds a threshold; and

responsive to determining that the similarity exceeds the threshold, selecting the first candidate root security query as the root security query.

18 . The computer system of claim 12 , wherein the computer-readable instructions are configured to cause the hardware processor to generate a requirement prompt to extract the security requirement associated with the root security query, the requirement prompt configured to query an indexed set of security requirements stored in a long-term memory of the ML generator.

19 . The computer system of claim 12 , wherein the computer-readable instructions are configured to cause the hardware processor to receive the system description input in a form of an image representation of the target system.

20 . Computer-readable storage media embodying computer readable instructions, the computer-readable instructions configured upon execution on a hardware processor to cause the hardware processor to:

receive a system description input pertaining a target system;

generate, using a machine learning (ML) generator, based on the system description input, a root security query;

extract from a security database a security requirement associated with the root security query;

output via a user interface an indication of the root security query;

receive first user input responsive to the indication of the root security query;

generate, using the ML generator, based on the first user input and the security requirement, a follow-up request;

output the follow-up request via the user interface;

receive second user input responsive to the follow-up request; and

based on the second user input, determine that the security requirement is not satisfied by the target system.