# US 2026/0094049 A1 · Identifying Flows in AI Algorithms

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0094049 A1
- Document ID: US20260094049A1
- Status: Published application
- Kind: A1
- Application number: 18/899,591
- Filed: 2024-09-27
- Published: 2026-04-02
- Earliest claimed: 2024-09-27
- Assignees: Micro Focus LLC
- Applicants: Micro Focus LLC
- Inventors: DOUGLAS MAX GROVER; MICHAEL F. ANGELO; MANISH MARWAH
- Web page: https://patentgrep.com/patent/US20260094049A1

## CPC

- G06N 20/00: Machine learning
- G06N 3/045: Combinations of networks
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Vectors of a target AI algorithm are captured by a trained flow AI algorithm. The trained flow AI algorithm is trained based on a training set of flows of vectors associated with a specific type of output data from the target AI algorithm. For example, the specific type of output data from the target AI algorithm may be a link to a malicious website. A flow of vectors in the target AI algorithm is identified by the trained flow AI algorithm based on the training set of flows of vectors associated with the specific type of output data from the target AI algorithm. In response to identifying the flow of vectors in the target AI algorithm, an action is taken. For example, the action may be to remove the link to the malicious website from the specific type of output data from the target AI algorithm.

## Claims

### Claim 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:
capture, by a trained flow Artificial Intelligence (AI) algorithm executed by the microprocessor, vectors of a target AI algorithm, wherein the trained flow AI algorithm is trained based on a training set of flows of vectors associated with a specific type of output data from the target AI algorithm;
identify, by the trained flow AI algorithm and based on the training set of flows of vectors associated with the specific type of output data from the target AI algorithm, a flow of vectors in the target AI algorithm; and
in response to identifying, based on the training set of flows of vectors associated with the specific type of output data from the target AI algorithm, the flow of vectors in the target AI algorithm, take an action.

### Claim 2 (depends on 1)

The system of claim 1, wherein the training set of flows of vectors associated with the specific type of output data are associated with a malicious type of output data.

### Claim 3 (depends on 1)

The system of claim 1, wherein the training set of flows of vectors associated with the specific type of output data also comprises a set of input prompts associated with the specific type of output data from the target AI algorithm.

### Claim 4 (depends on 1)

The system of claim 1, wherein the action is at least one of: modifying the specific type of output data from the target AI algorithm, blocking the specific type of output data from the target AI algorithm, modifying an input prompt to the target AI algorithm, blocking an input prompt to the target AI algorithm, stopping the target AI algorithm from completing processing of an input prompt, and identifying a source of an input prompt to the target AI algorithm.

### Claim 5 (depends on 1)

The system of claim 1, wherein the flow of vectors in the target AI algorithm comprises at least one of:
a flow of vectors from an input node to an output node;
a flow of vectors from internal node to an output node;
a flow of vectors from an input node to an internal node;
a flow of vectors from an internal node to an internal node;
a flow of vectors from an input node through two or more parallel nodes;
a flow of vectors from an internal node through two parallel nodes; and
a plurality of flows of vectors with a plurality of separate paths.

### Claim 6 (depends on 1)

The system of claim 1, wherein identifying the flow of vectors in the target AI algorithm is based on capturing a subset of vectors of the target AI algorithm.

### Claim 7 (depends on 6)

The system of claim 6, wherein the captured subset of vectors of the target AI algorithm are identified based on at least one of: specific flows between specific layers, flows related to specific tokens, specific identified flows, a binary classifier, a dimensionality, a Principal Component Analysis (PCA), and an autoencoder.

### Claim 8 (depends on 1)

The system of claim 1, wherein the action is to check to see if any weights associated with the flow of vectors in the target AI algorithm have been changed.

### Claim 9 (depends on 1)

The system of claim 1, wherein capturing the vectors of the target AI algorithm is done in real-time.

### Claim 10

A method comprising:
capturing, by a trained flow Artificial Intelligence (AI) algorithm, vectors of a target AI algorithm, wherein the trained flow AI algorithm is trained based on a training set of flows of vectors associated with a specific type of output data from the target AI algorithm;
identifying, by the trained flow AI algorithm and based on the training set of flows of vectors associated with the specific type of output data from the target AI algorithm, a flow of vectors in the target AI algorithm; and
in response to identifying, based on the training set of flows of vectors associated with the specific type of output data from the target AI algorithm, the flow of vectors in the target AI algorithm, taking an action.

### Claim 11 (depends on 10)

The method of claim 10, wherein the training set of flows of vectors associated with the specific type of output data are associated with a malicious type of output data.

### Claim 12 (depends on 10)

The method of claim 10, wherein the training set of flows of vectors associated with the specific type of output data also comprises a set of input prompts associated with the specific type of output data from the target AI algorithm.

### Claim 13 (depends on 10)

The method of claim 10, wherein the action is at least one of: modifying the specific type of output data from the target AI algorithm, blocking the specific type of output data from the target AI algorithm, modifying an input prompt to the target AI algorithm, blocking an input prompt to the target AI algorithm, stopping the target AI algorithm from completing processing of an input prompt, and identifying a source of an input prompt to the target AI algorithm.

### Claim 14 (depends on 10)

The method of claim 10, wherein the flow of vectors in the target AI algorithm comprises at least one of:
a flow of vectors from an input node to an output node;
a flow of vectors from internal node to an output node;
a flow of vectors from an input node to an internal node;
a flow of vectors from an internal node to an internal node;
a flow of vectors from an input node through two or more parallel nodes;
a flow of vectors from an internal node through two parallel nodes; and
a plurality of flows of vectors with a plurality of separate paths.

### Claim 15 (depends on 10)

The method of claim 10, wherein identifying the flow of vectors in the target AI algorithm is based on capturing a subset of vectors of the target AI algorithm.

### Claim 16 (depends on 15)

The method of claim 15, wherein the captured subset of vectors of the target AI algorithm are identified based on at least one of: specific flows between specific layers, flows related to specific tokens, specific identified flows, a binary classifier, a dimensionality, a Principal Component Analysis (PCA), and an autoencoder.

### Claim 17 (depends on 10)

The method of claim 10, wherein the action is to check to see if any weights associated with the flow of vectors in the target AI algorithm have been changed.

### Claim 18 (depends on 10)

The method of claim 10, wherein capturing the vectors of the target AI algorithm is done in real-time.

### Claim 19

A non-transient computer readable medium having stored thereon instructions that cause a microprocessor to execute a method, the method comprising instructions to:
capture, by a trained flow Artificial Intelligence (AI) algorithm executed by the microprocessor, vectors of a target AI algorithm, wherein the trained flow AI algorithm is trained based on a training set of flows of vectors associated with a specific type of output data from the target AI algorithm;
identify, by the trained flow AI algorithm and based on the training set of flows of vectors associated with the specific type of output data from the target AI algorithm, a flow of vectors in the target AI algorithm; and
in response to identifying, based on the training set of flows of vectors associated with the specific type of output data from the target AI algorithm, the flow of vectors in the target AI algorithm, take an action.

### Claim 20 (depends on 19)

The non-transient computer readable medium of claim 19, wherein identifying the flow of vectors in the target AI algorithm is based on capturing a subset of vectors of the target AI algorithm.
