Granted Patent B1

US 12,488,024 · App. 18/986,122 · Granted Dec 2, 2025

Inventors: Leonid Feinberg (London, GB); Oren Gev (Plano, TX); Ishai Rosenberg (Modiin, IL)

Assignee: Verax AI Trust LTD

G06F16/285G06F16/24565

Abstract

Claims (90)

1 . A method for applying controls on an input for an artificial intelligence model:

receiving an input for a system having a generative artificial intelligence (AI) model, wherein the generative AI model is configured to generate an output based on at least a data source;

determining at least a constraint for the input based on a predetermined constraint from a plurality of preexisting constraints, including a retrieval augmented generation (RAG);

generating a modified input based on the determined constraint;

processing the modified input utilizing the generative artificial AI model to generate an output;

generating a first contextual value based on the modified input;

generating a second contextual value based on the generated output;

generating a score based on any one of: the first contextual value, the second contextual value, and a combination thereof; and

sending the generated output to a client of the system.

2 . The method of claim 1 , further comprising:

generating the predetermined constraint based on an analysis of a plurality of input-output pairs, each input-output pair including an output generated by the generative AI based on an input.

3 . The method of claim 2 , further comprising:

clustering the plurality of input-output pairs into a plurality of input-output groups, wherein each input-output group is clustered based on the input.

4 . The method of claim 2 , further comprising:

clustering the plurality of input-output pairs into a plurality of input-output groups, wherein each input-output group is clustered based on the output.

5 . The method of claim 2 , wherein the analysis of the plurality of input-output pairs further comprises:

generating a verification score for each output of an input-output pair; and

generating the predetermined constraint based on input-output pairs having a verification score which satisfies condition based on a predetermined threshold.

6 . The method of claim 5 , further comprising:

detecting at least an attribute common to each input of the input-output pairs having the verification score which satisfies the condition; and

generating the modified input further based on the detected attribute.

7 . The method of claim 6 , further comprising:

detecting at least another attribute common to each input of the input-output pairs having a verification score which does not satisfy the condition; and

generating the modified input further based on the detected at least another attribute.

8 . The method of claim 1 , further comprising:

generating the predetermined constraint based on an analysis of a plurality of inputs.

9 . The method of claim 1 , further comprising:

generating the predetermined constraint based on an analysis of a plurality of outputs.

10 . The method of claim 1 , further comprising:

sending the generated output in response to determining that the verification score satisfies a condition based on a predetermined threshold.

11 . The method of claim 10 , further comprising:

initiating generation of a second modified input, in response to determining that the verification score does not satisfy a condition;

processing the second modified input to generate another output; and

sending the another output in response to determining that a verification score based on the another output satisfies the condition.

12 . The method of claim 11 , further comprising:

generating the second modified input based on a second determined constraint.

13 . The method of claim 1 , further comprising:

generating a metadata based on the verification score; and

sending the generated output and the generated metadata.

14 . The method of claim 1 , further comprising:

generating the verification score further based on data extracted from the at least a data source.

15 . The method of claim 1 , wherein sending the output further comprises:

modifying the output based on any one of: sanitizing the output, generating a semantically compressed output, modifying a tone of the output, and any combination thereof.

16 . A non-transitory computer-readable medium storing a set of instructions for applying controls on an input for an artificial intelligence model, the set of instructions comprising:

one or more instructions that, when executed by one or more processing circuitries of a device, cause the device to:

receive an input for a system having a generative artificial intelligence (AI) model, wherein the generative AI model is configured to generate an output based on at least a data source;

determine at least a constraint for the input based on a predetermined constraint from a plurality of preexisting constraints, including a retrieval augmented generation (RAG);

generate a modified input based on the determined constraint;

process the modified input utilizing the generative artificial AI model to generate an output;

generate a first contextual value based on the modified input;

generate a second contextual value based on the generated output;

generate a score based on any one of: the first contextual value, the second contextual value, and a combination thereof; and

send the generated output to a client of the system.

17 . A system for applying controls on an input for an artificial intelligence model: comprising:

one or more processing circuitry circuitries configured to:

receive an input for a system having a generative artificial intelligence (AI) model, wherein the generative AI model is configured to generate an output based on at least a data source;

determine at least a constraint for the input based on a predetermined constraint from a plurality of preexisting constraints, including a retrieval augmented generation (RAG);

generate a modified input based on the determined constraint;

process the modified input utilizing the generative artificial AI model to generate an output;

generate a first contextual value based on the modified input;

generate a second contextual value based on the generated output;

generate a score based on any one of: the first contextual value, the second contextual value, and a combination thereof; and

send the generated output to a client of the system.

18 . The system of claim 17 , wherein the one or more processing circuitries circuitry are further configured to:

generate the predetermined constraint based on an analysis of a plurality of input-output pairs, each input-output pair including an output generated by the generative AI based on an input.

19 . The system of claim 18 , wherein the one or more processing circuitries circuitry are further configured to:

cluster the plurality of input-output pairs into a plurality of input-output groups, wherein each input-output group is clustered based on the input.

20 . The system of claim 18 , wherein the one or more processing circuitries circuitry are further configured to:

cluster the plurality of input-output pairs into a plurality of input-output groups, wherein each input-output group is clustered based on the output.

21 . The system of claim 18 , wherein the analysis of the plurality of input-output pairs further comprises:

generating a verification score for each output of an input-output pair; and

generating the predetermined constraint based on input-output pairs having a verification score which satisfies condition based on a predetermined threshold.

22 . The system of claim 21 , wherein the one or more processing circuitries circuitry are further configured to:

detect at least an attribute common to each input of the input-output pairs having the verification score which satisfies the condition; and

generate the modified input further based on the detected attribute.

23 . The system of claim 22 , wherein the one or more processing circuitries circuitry are further configured to:

detect at least another attribute common to each input of the input-output pairs having a verification score which does not satisfy the condition; and

generate the modified input further based on the detected at least another attribute.

24 . The system of claim 17 , wherein the one or more processing circuitries circuitry are further configured to:

generate the predetermined constraint based on an analysis of a plurality of outputs.

25 . The system of claim 17 , wherein the one or more processing circuitries circuitry are further configured to:

send the generated output in response to determining that the verification score satisfies a condition based on a predetermined threshold.

26 . The system of claim 25 , wherein the one or more processing circuitries circuitry are further configured to:

initiate generation of a second modified input, in response to determining that the verification score does not satisfy a condition;

process the second modified input to generate another output; and

send the another output in response to determining that a verification score based on the another output satisfies the condition.

27 . The system of claim 26 , wherein the one or more processing circuitries circuitry are further configured to:

generate the second modified input based on a second determined constraint.

28 . The system of claim 17 , wherein the one or more processing circuitries circuitry are further configured to:

generate the verification score further based on data extracted from the at least a data source.

Assignments (1)

References Cited (15)