Abstract: The technology disclosed includes a system to group security alerts generated in a computer network and prioritize grouped security alerts for analysis, through graph-based clustering. The graph used to form clusters includes entities in the computer network represented as scored nodes, and relationships of entities as weighted edges. The technology disclosed includes traversing the graph starting at starting nodes and propagating native scores through and to neighboring nodes connected by the weighted edges. The propagated scores at visited nodes are normalized by attenuation based on contributing neighboring nodes of a respective visited node. An aggregate score for a visited node is calculated by accumulating propagated scores at visited nodes with their respective native scores. The technology disclosed forms clusters of connected nodes in the graph that have a respective aggregate score above a selected threshold. The clusters are ranked and prioritized for analysis, pursuant to the aggregate scores.

Abstract: Disclosed is a method of building a customized deep learning (DL) stack classifier to detect organization sensitive data in images, referred to as image-borne organization sensitive documents, and protecting against loss of the image-borne organization sensitive documents, including distributing a trained feature map extractor stack, with stored parameters, configured to allow the organization to extract from image-borne organization sensitive documents, feature maps that are used to generate updated DL stacks and to save non invertible feature maps derived from the images, and ground truth labels for the image. Also included is receiving organization-specific examples including the non-invertible feature maps extracted from the organization-sensitive documents and the ground truth labels and using the received organization-specific examples to update a customer-specific DL stack classifier. Further included is sending the customer-specific DL stack classifier to the organization.

Abstract: The technology disclosed teaches protecting sensitive data in the cloud via indexable databases. The method includes identifying sensitive fields of metadata for encryption and for hashing. The method also includes hashing at least partial values in the indexable sensitive fields to non-reversible hash values, concatenating the non-reversible hash values with the metadata for the network events, and encrypting the sensitive fields of metadata. Also included is sending the metadata for the network events, with the non-reversible hash values and the encrypted sensitive fields, to a remote database server that does not have a decryption key for the encrypted sensitive fields and that indexes the non-reversible hash values for indexed retrieval against the indexable sensitive fields.

Abstract: The technology disclosed relates to a method, system, and non-transitory computer-readable media that trains a cloud traffic classifier to classify cross-application communications as malicious command and control (C2) traffic or benign cloud traffic. The training uses blocks of malicious Hypertext Transfer Protocol (HTTP) transactions targeted at a plurality of cloud applications by a plurality of clients prequalified as malicious command and control (C2) cloud traffic, and also blocks of benign HTTP transactions targeted at the plurality of cloud applications by the plurality of clients prequalified as benign cloud traffic. A cloud traffic classifier is trained on the cross-application malicious training example set and on the cross-application benign training example set by processing the blocks of the malicious and benign HTTP transactions as inputs, and generating outputs that classify the training examples as respectively malicious C2 cloud traffic or benign cloud traffic.

Abstract: A policy-based security system for establishing a secure session from client devices to a web server includes a policy component with policies, a client device with a local application to select a cloud service, and a mid-link server. A set of policies from the policies is determined. An encryption link specified for the set of policies and the cloud service is determined. A set of session protocols is selected to establish the secure session between the client device and the web server based on the set of policies. It is determined whether the client device satisfies security standards of one or more session protocols from the set and based on the determination, either a direct link is selected to establish the secure session using a session protocol from the set or a secure tunnel between the client device and the mid-link server and a corresponding tunnel protocol is selected.

Abstract: A system for providing policy-controlled communication over the Internet includes a client endpoint function that executes on a client device while coupled to a first VPN tunnel, a service endpoint function that operates a remote service of a plurality of remote services, and a mid-link server coupled to the first VPN tunnel and a second VPN tunnel. The client endpoint function includes a first VPN endpoint component, and the service endpoint function includes a second VPN endpoint component. A router component operates to route network packet traffic between the first and second VPN tunnels via a route specified by a plurality of policies, an inspection component that analyzes network packet traffic in accordance with the plurality of policies. The plurality of policies for the network packet traffic and the content mediation selected dynamically on the basis of one or more of a user, an application, an endpoint, and a session.

Abstract: Disclosed is phishing classifier that classifies a URL and content page accessed via the URL as phishing or not is disclosed, with URL feature hasher that parses and hashes the URL to produce feature hashes, and headless browser to access and internally render a content page at the URL, extract HTML tokens, and capture an image of the rendering. Also disclosed are an HTML encoder, trained on HTML tokens extracted from pages at URLs, encoded, then decoded to reproduce images captured from rendering, that produces an HTML encoding of the tokens extracted, and an image embedder, pretrained on images, that produces an image embedding of the image captured. Further, phishing classifier layers, trained on the feature hashes, the HTML encoding, and the image embedding, process the URL feature hashes, HTML encoding and image embeddings to produce a likelihood score that the URL and the page accessed presents a phishing risk.

Abstract: The technology disclosed includes a system to efficiently classify sensitivity of document generated by and downloaded from cloud-based provider services. The system monitors network traffic at a document-generation initiating endpoint and receives a web page identifying the document generated. The system parses the network traffic that selects the document for download, based on the user selecting a link, and intercepts a document handle in an API parameter string used to download the document. The system interprets the document handle to analyze sensitivity of the document to assign a sensitive classification to the document. The sensitivity classification is encoded into the document header metadata. The encoded sensitivity classification can be used to enhance security, for example, preventing data exfiltration.

Abstract: Systems and methods for providing policy-controlled communication over the Internet are provided. A system may include a client endpoint function configured to execute on a client device while coupled to a first VPN tunnel, a service endpoint function that operates a remote service of a plurality of remote services, a gateway server including a first VPN termination point that authenticates and terminates the first VPN tunnel, a stitcher server including a second VPN termination point that authenticates and terminates a second VPN tunnel, and a mid-link server coupled to the first VPN tunnel and the second VPN tunnel. The mid-link server may include a plurality of Access Resource Servers (ARSs), and the gateway server and the stitcher server may communicate via a network connecting the plurality of ARSs.

Abstract: An automatically predetermined credential system for a remote administrative operating system (OS) authorization and policy control is disclosed. Administrative activities are packaged in single-use downloaded software program. When executed, the administrative access to the OS is activated before completing the administrative activities. The single-use downloaded software program has policies that performs checks on a user computer executing the software program. The policies include checking firewall settings, confirming virus checking, interrogating software to confirm patches or updates have been performed, checking for key loggers or other surveillance software or devices The single-use downloaded software is protected with a passcode to prevent activation in an unauthorized way.

Abstract: A method and system for policy-driven traffic management in cloud-based multi-tenant systems is disclosed. Routes delivers services to end user devices. The routes are specified for policies. The policies specify residencies for the routes, cloud services, and data storage. An application running on an end user device selects a policy and a cloud service. A route corresponding to the policy, a residency specified by the policy, and the cloud service is returned to the application. An operation of the end user device in compliance with the policies is monitored to generate telemetry. An access resource server receives a non-compliance with the policy, and identify new routes, policies and configuration for remediation.

Abstract: A key management system for providing encryption of a disk in a client device is provided. The system comprises a trusted platform module (TPM) having a first fragment of a key, a remote storage having a second fragment of the key, and a processing unit to partially boot instructions relating to booting of the client device, send a request for validation of the to the TPM, receive the first fragment of the key from the TPM on successful validation, request for the second fragment of the key with credentials to access the remote storage. The credentials and a network of the request are verified, the second fragment of the key is transmitted on successful validation. The first fragment and the second fragment of the key are combined to generate an encryption key for booting the client device. The first fragment of the key and the second fragment of the key are rotatable.

Abstract: The technology disclosed applies data loss prevention (DLP) to those cloud-applications for which no application-specific parser is available. Known cloud applications can be arranged in categories of services such as “personal pages and blog,” “news websites,” “cloud-based storage services,” and “social media services.” A category includes a list of uniform resource locators (URLs) of providers of cloud applications that allow users to perform similar activities. The various providers in a category use different syntaxes to implement services in the category. The disclosed category-directed parsers synthesize interaction syntax patterns of a sample of providers in the category. A category-directed parser collects metadata from known cloud applications using multiple category-directed match rules synthesized from syntaxes used by the sample providers in the category.

Abstract: Disclosed is detecting identification documents in image-borne identification documents and protecting against loss of the image-borne identification documents. A trained deep learning (DL) stack is used to classify production images by inference as containing a sensitive image-borne identification document, with the trained stack configured with parameters determined using labelled ground truth data for the identification documents and examples of other image documents. The trained DL stack is configured to include a first set of layers closer to an input layer and a second set of layers further from the input layer, with the first set pre-trained to perform image recognition before exposing the second set of layers of the stack to the labelled ground truth data for the image-borne identification documents and examples of other image documents, and using the inferred classification of the sensitive image-borne identification document in a DLP system to protect against loss by image exfiltration.

Abstract: A method and system for reducing triggering of throughput penalties imposed on a group of users by a software-as-a-service (SaaS) server due to Application Programming Interface (API) calls exceeding limits of the SaaS server. The approaches include intercepting requests to the SaaS server from a user group and monitoring both a rate of API calls for the requests and a rate of API events generated by forwarding the API calls to the SaaS server, intercepting the SaaS server's responses, where some of the responses indicate a throughput penalty imposed by the server, identifying one or more power users from the user group based on the rate of generated notifications, and throttling the rate of the API calls for the requests submitted by the identified power users of the user group to the SaaS server in accordance with an API call throttle limit, thus reducing triggering of the throughput penalty.

Abstract: Systems and methods to generate a device composite fingerprint associated with a computing device are described. In one embodiment, communication data associated with the computing device is accessed. The communication data includes device identification data, device group data, and device operational data. A device identity fingerprint associated with the computing device is generated using the device identification data. A device group fingerprint associated with the computing device is generated using the device group data. A device operational fingerprint associated with the computing device is generated using the device operational data. The device identity fingerprint, the device group fingerprint, and the device operational fingerprint are combined to generate a device composite fingerprint.

Abstract: A policy-controlled access security system for managing access security to electronic agents in cloud based multi-tenant systems includes a client device, a mid-link server, and a web server. A local application running on the client device requests for access to an electronic agent of a remote application of the web server. Policies are determined corresponding to the electronic agent for controlled access to the electronic agent. A token for the electronic agent is correlated with a plurality of tokens for identifying a user application associated with the token. The remote application corresponding to the token from the request is compared with the user application. Based on the comparison an authorization is determined by the mid-link server corresponding to the token for accessing the electronic agent. The policies are enforced on the client device and the access to the electronic agent is provided based on the policies via the web server.

Abstract: A policy-controlled authorization system for managing tokens used to access services in a cloud based multi-tenant system. The policy-controlled authorization system includes a local application that executes on a client device, a policy component including a plurality of policies, and a mid-link server, coupled to the client device. A request for access to a service on a remote application running on a remote instance of a web server is provided by the local application. A token is required to access the service. A correlator correlates the token with the plurality of tokens for identifying a policy from the plurality of policies associated with the token. A token inspector authorizes the token for accessing the service based on the correlation. Based on the authorization, either the token is authorized for access to the service via the remote application, or the token is blocked when unauthorized to prevent access to the service.

Abstract: The technology disclosed relates to distributing a trained master deep learning (DL) stack with stored parameters to a plurality of organizations, to detect organization sensitive data in images, referred to as image-borne organization sensitive documents, and protecting against loss of the image-borne organization sensitive documents. Disclosed is providing organizations with a DL stack update trainer, under the organizations' control, configured to allow the organizations to perform update training to generate updated DL stacks, without the organizations forwarding images of organization-sensitive training examples, and to save non-invertible features derived from the images, ground truth labels for the images, and parameters of the updated DL stacks.

Abstract: Systems and methods for providing policy-controlled communication over the Internet are provided. A system may include a client endpoint function configured to execute on a client device while coupled to a first VPN tunnel, a service endpoint function that operates a remote service of a plurality of remote services, and a mid-link server coupled to the first VPN tunnel and a second VPN tunnel. The mid-link server may include an inspection component that analyzes network packet traffic in accordance with a plurality of policies. The inspection component may inspect the network packet traffic for specific content and provide instructions to a router component and/or a mediation component of the mid-link server. The instructions may be a function of at least one policy that applies to the specific content.