Abstract: A method of identifying malicious activity in a sequence of computer instructions includes providing the sequence of computer instructions into a recurrent neural network configured to provide an output based on both the current instruction being input and at least one prior instruction in the sequence, and evaluating the provided sequence of computer instructions in the recurrent neural network at multiple points within the sequence. An output is provided indicating whether the network has determined the code sequence to that point is likely malicious.

Abstract: A third-party attempting to call a user is identified and authenticated by collecting multi-dimensional user data and user activity into a personalized large language model (LLM) residing on a user device. A general-purpose LLM generates third-party authentication questions based on an initial intent of the caller and the questions are provided to the personalized LLM with the answers from the personalized LLM ranked based on relevancy and accuracy. One or more of the highest ranked questions are prompted to the caller and the answers provided by the caller are compared to the answers provided by the personalized LLM to determine if the caller is a trusted contact or a likely malicious third-party.

Abstract: Differentiating between human interaction and computer interaction using a natural perception of human eyes to identify human users. One or more images may be displayed within one or more test fields on a web page. The one or more images themselves, or a duration for which they are presented, may cause a human eye to not see something that is in the images or to see something that is not in the images. Input may be received from the user identifying all images that are perceived by the user. A determination may be made that the user is a human user based on what the user identifies. Access to another web page may be granted to users who are determined to be human and denied to users that are not determined to be human.

Abstract: In one embodiment, a method may include receiving a first input query from a client device, generating a first set of related queries based on the first input query, executing a search to generate a first response to the first input query, storing the first response, the first input query, and the first set of related queries in a cache such that the first response is associated with the first input query and the first set of related queries, and providing, in response to receiving the first input query, the first response to the client device.

Abstract: A computer user is authenticated by collecting user data from at least two dimensions of user activity into a local large language model (LLM) residing on a user device, and generating user authentication questions from at least those two dimensions of user activity using the large language model. Questions having intersecting time or location are discarded, and questions from two or more dimensions are presented to the user for response. The user is authenticated by comparing the user response to known correct answers to the questions.

Abstract: Obfuscation techniques used in phishing attacks that evade detection using current machine learning models can be identified and statistically characterized for use in generating adversarial training samples to improve the performance of a machine learning model phishing classifier. By using statistical information regarding the prevalence and effectiveness of various obfuscation techniques observed in real-world samples, adversarial samples that are effective and realistic examples of phishing content that evades phishing content classifier detection can be generated and used to augment the data set used for training or re-training the machine learning model classifier.

Abstract: A method for accessing a network resource including detecting an attempt by a user via a computing device to access a service enabled by a computing system via a network and transmitting via the network to the computing system a first request to access the service in response to detecting the attempt by the user to access the service, the first request including at least one empty personally identifiable data structure. A failure to access the service responsive to the first request is determined. A second request to access the service in response to the first failure to access the service is transmitted via the network to the computing system, the second request including artificial personally identifiable information, and access to the service from the computing system is received for the user.

Abstract: Systems and methods for transacting over a network. A first agent and a second agent are provided. The second agent is operable to transact with a third agent for use of a network-enabled service based on a first transaction policy from a fourth agent, the third agent enabled to communicate with a fifth agent. The first agent is operable to communicate with the second agent to facilitate the transacting by the second agent with the third agent for the use of the network-enabled service based on the first transaction policy and communicate with the fifth agent to facilitate the transacting by the second agent with the third agent for the use of the network-enabled service.

Abstract: Systems and methods for classifying user intent based on the user interaction with an entry bar on a user device include receiving an input from a user in an entry box; classifying the input into a category of a plurality of categories including a category for an Artificial Intelligence (AI) session for an AI agent; and performing an action responsive to the classified category, including utilizing the AI agent for the AI session when the category is the category for the AI session and bypassing the AI agent for other categories of the plurality of categories.

Abstract: A malware classification system provides improved confidence in explanations of neural network classification outputs using methods such as weighting or masking when training the neural network to train the network on a sample resembling or including the explanation. The explanation in some examples comprises a subset of a hierarchical input vector that is responsible for the neural network's classification output. In another example the neural network has an inner portion configured to reduce the weight of elements of the output not significantly contributing to the explanation of the output, such as by reducing the weight of as many such outputs to zero as is practical in generating the desired output.

Abstract: A method and system detects at a plurality of network locations a plurality of accuracy ratings of a plurality of media instances and detects the plurality of media instances. A particular accuracy rating of one or more particular media instances is detected at a particular network location, and the one or more particular media instances are detected. A bias of the particular accuracy rating is determined based on the particular accuracy rating, the one or more particular media instances, the plurality of accuracy ratings, and the plurality of media instances. An indication is transmitted to a user based on the bias of the particular accuracy rating.

Abstract: Systems and methods for in-transit email review and data collection. In one embodiment, a method may include receiving a plurality of original emails at an alias email address. The alias email address may be associated with a user. Once a plurality of original emails are received at the alias email address, the method may include analyzing the plurality of original emails to identify relevant data contained within the plurality of original emails, and storing a copy of the identified relevant data in a database that is accessible by the user.

Abstract: Systems and methods for transacting over a network is provided. The system includes a first agent and second agent. The first agent is operable to receive from a third agent a transaction code associated with one or more credential types required to apply the transaction code or with one or more credential claim types required to apply the transaction code, transmit the transaction code to the second agent, and receive from the second agent a digitally signed transaction, a first verifiable proof, and the transaction code. The first agent is further operable to transmit to a fourth agent a second verifiable proof based on the first verifiable proof and the transaction code, receive from the fourth agent an unlock signature for a locked credential including one or more credential claims, and transmit the unlock signature to the second agent.

Abstract: A system and method are provided by which an electronic address associated with a user is monitored. Based on the monitoring, an electronic message is detected including a digital document. A cryptographic function is applied to the digital document to generate a hash which is rendered accessible at a network location. An identification of the network location of the hash is transmitted to a first computing system associated with the user.

Abstract: Systems and methods for Artificial Intelligence (AI) model bundling and splitting for widescale distribution include obtaining an Artificial Intelligence (AI) model having a plurality of layers; producing a plurality of slices, each slice is a layer and an associated encoding of the layer; determining a header and a trailer each defining a previous slice and a next slice, respectively, and optimization coefficients; and serving the plurality of slices and corresponding headers, trailers, and optimization coefficients.

Abstract: Systems and methods for classifying user intent based on the user interaction with an entry bar on a user device include receiving an input from a user in an entry box; classifying the input into a category of a plurality of categories including a category for an Artificial Intelligence (AI) session for an AI agent; and performing an action responsive to the classified category, including utilizing the AI agent for the AI session when the category is the category for the AI session and bypassing the AI agent for other categories of the plurality of categories.

Abstract: Systems and methos for using general-purpose Artificial Intelligence (AI) models as special purpose classifiers include utilizing a prompt for a general-purpose model with the prompt including instructions to perform a classification of an input into one of a plurality of categories; tokenizing the prompt and the input into a plurality tokens including a unique token for each of the plurality of categories; biasing weights of the plurality of tokens such that the unique token for each of the plurality of categories have greater weights than other tokens of the plurality of tokens; and inputting the plurality of tokens with their corresponding weights into the AI model.

Abstract: Systems and methods are provided as an AI-powered web browser companion. A method, according to one implementation, includes a step of locally monitoring user interactions during a web browsing session associated with a client device operating within a browser environment. The method also includes a step of locally determining user interests, user preferences, and behavioral patterns based on the monitored user interactions and further based on previously stored web browsing sessions. Furthermore, the method includes a step of providing personalized content and predictive navigation assistance in the browser environment based on the user interests, user preferences, and behavioral patterns.

Abstract: Systems and methods are provided as a web page management tool. A method, according to one implementation, includes a step of analyzing a plurality of browsing pages that are currently open and active on a Graphical User Interface (GUI), each browsing page related to a user browsing session. The method also includes a step of employing a Machine Learning (ML) model to determine when to close one or more pages of the plurality of browsing pages based on predefined conditions. Before closing the one or more pages, the method also includes a step of archiving metadata and at least a portion of content related to each of the one or more pages to thereby enable a user to retrieve at least a portion of the one or more pages at a later time.

Abstract: Systems and methods for maintaining user privacy in reverse personal information searches are disclosed. A user seeking to search a network (such as the Internet) for instances of one or more digital assets belonging to the user uploads the digital assets to the server system where they are stored in a secure, encrypted storage. A threshold value defines how similar a located digital asset must be to an original digital asset in order to be considered a match. The server system assembles queries to search for unauthorized instances of the user's digital assets and performs the search. Located assets and their locations (URL, IP address, etc.) are stored for access and viewing by the authorized user. Uploaded digital assets and search results are encrypted so that the operator of the server system cannot view the original digital assets or the search results to further protect the authorized user's privacy.