# US 2026/0254836 A1 · Detection and prevention of external fraud

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

- Number: US 2026/0254836 A1
- Document ID: US20260254836A1
- Status: Published application
- Kind: A1
- Application number: 19/422,210
- Filed: 2025-12-16
- Published: 2026-08-27
- Earliest claimed: 2020-04-23
- Assignees: Abnormal AI, Inc.
- Applicants: Abnormal AI, Inc.
- Inventors: Yu Zhou Lee; Lawrence Stockton Moore; Jeshua Alexis Bratman; Lei Xu; Sanjay Jeyakumar
- Web page: https://patentgrep.com/patent/US20260254836A1

## CPC

- H04L 63/1433: {Vulnerability analysis}
- G06F 40/30: Semantic analysis
- H04L 51/08: Annexed information, e.g. attachments
- H04L 51/212: using filtering or selective blocking
- H04L 63/0236: {Filtering by address, protocol, port number or service, e.g. IP-address or URL}
- H04L 63/0245: {Filtering by information in the payload}
- H04L 63/101: {Access control lists [ACL]}
- H04L 63/126: {the source of the received data}
- H04L 63/145: {the attack involving the propagation of malware through the network, e.g. viruses, trojans or worms}
- H04L 63/1483: {service impersonation, e.g. phishing, pharming or web spoofing (detection of rogue wireless access points H04W12/12)}
- H04L 51/18: Commands or executable codes
- H04L 51/222: using geographical location information, e.g. messages transmitted or received in proximity of a certain spot or area
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Techniques for detecting instances of external fraud by monitoring digital activities that are performed with accounts associated with an enterprise are disclosed. In one example, a threat detection platform determines the likelihood that an incoming email is indicative of external fraud based on the context and content of the incoming email. To understand the risk posed by an incoming email, the threat detection platform may seek to determine not only whether the sender normally communicates with the recipient, but also whether the topic is one normally discussed by the sender and recipient. In this way, the threat detection platform can establish whether the incoming email deviates from past emails exchanged between the sender and recipient.

## Claims

### Claim 1

A system comprising:
a processor configured to:
obtain a first set of emails received by employees of an enterprise;
examine content of the first set of emails so as to identify a second set of emails that contain invoices, wherein the second set of emails is a subset of the first set of emails;
receive input indicative of labels for the second set of emails that are provided through an interface, wherein each label indicates whether the corresponding email is representative of a legitimate request for payment; and
train a model to identify legitimate requests for payment by providing (i) the second set of emails and (ii) the labels to the model as training data; and
a memory coupled to the processor and configured to provide the processor with instructions.

### Claim 2 (depends on 1)

The system of claim 1, wherein the processor is further configured to:
perform a resolution procedure so as to identify email addresses from which the first set of emails are received; and
establish that each email included in the second set of emails is transmitted by a vendor based on the corresponding email address.

### Claim 3 (depends on 1)

The system of claim 1, wherein the processor is further configured to apply heuristic functions to the first set of emails to identify the second set of emails that contain invoices.

### Claim 4 (depends on 1)

The system of claim 1, wherein the interface includes an entity labeler platform configured to allow manual labeling of the second set of emails.

### Claim 5 (depends on 1)

The system of claim 1, wherein the labels includes human-defined labels that establish a ground truth dataset for training the model.

### Claim 6 (depends on 1)

The system of claim 1, wherein the processor is further configured to: obtain an incoming email; and apply the trained model to the incoming email to determine a likelihood that the incoming email includes a request for payment.

### Claim 7 (depends on 6)

The system of claim 6, wherein applying the trained model to the incoming email comprises analyzing at least one of a subject, a body, or an attachment of the incoming email to determine whether the incoming email contains an invoice.

### Claim 8 (depends on 6)

The system of claim 6, wherein applying the trained model to the incoming email comprises analyzing text included in a body of the incoming email using a bag-of-words model.

### Claim 9 (depends on 6)

The system of claim 6, wherein applying the trained model to the incoming email comprises determining whether an attachment of the incoming email contains a dollar sign or a physical address.

### Claim 10 (depends on 6)

The system of claim 6, wherein applying the trained model to the incoming email comprises analyzing terms contained in an invoice attached to the incoming email using optical character recognition or text analysis.

### Claim 11 (depends on 6)

The system of claim 6, wherein the processor is further configured to perform a vendor resolution procedure to identify a vendor involved in the incoming email.

### Claim 12 (depends on 11)

The system of claim 11, wherein performing the vendor resolution procedure comprises applying a natural language processing algorithm or a computer vision algorithm to a body of the incoming email to establish an identity of the vendor.

### Claim 13 (depends on 11)

The system of claim 11, wherein performing the vendor resolution procedure comprises fetching records of domain age from a database to determine an age of a domain from which the incoming email originated.

### Claim 14 (depends on 6)

The system of claim 6, wherein the processor is further configured to: determine, based on the trained model, that the incoming email is representative of a request to pay an invoice; and upload the incoming email to a vendor database representing a record of counts built off a history of invoice-related conversations.

### Claim 15 (depends on 14)

The system of claim 14, wherein the processor is further configured to generate a metric indicative of risk for the incoming email based on a degree of similarity between the incoming email and past emails for which information is available in the vendor database.

### Claim 16 (depends on 15)

The system of claim 15, wherein the processor is further configured to implement a filter to at least temporarily prohibit emails from a vendor associated with the incoming email responsive to the metric.

### Claim 17 (depends on 16)

The system of claim 16, wherein the processor is further configured to cause display of a notification including information regarding the incoming email responsive to the metric indicating that a risk in communicating with a vendor associated with the incoming email is high.

### Claim 18 (depends on 1)

The system of claim 1, wherein the processor is further configured to include synthetic emails in the training data, wherein the synthetic emails are not actually sent by a sender to a recipient.

### Claim 19

A method comprising:
obtaining a first set of emails received by employees of an enterprise;
examining content of the first set of emails so as to identify a second set of emails that contain invoices, wherein the second set of emails is a subset of the first set of emails;
receiving input indicative of labels for the second set of emails that are provided through an interface, wherein each label indicates whether the corresponding email is representative of a legitimate request for payment; and
training a model to identify legitimate requests for payment by providing (i) the second set of emails and (ii) the labels to the model as training data.

### Claim 20

A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:
obtaining a first set of emails received by employees of an enterprise;
examining content of the first set of emails so as to identify a second set of emails that contain invoices, wherein the second set of emails is a subset of the first set of emails;
receiving input indicative of labels for the second set of emails that are provided through an interface, wherein each label indicates whether the corresponding email is representative of a legitimate request for payment; and
training a model to identify legitimate requests for payment by providing (i) the second set of emails and (ii) the labels to the model as training data.
