Sap Se

Sap Ag

Slim Trabelsi

Laurent Y Gomez

Gilles Montagnon

Rocio Cabrera Lozoya

Marco Rosa

Carlos Rafael Ocanto Davila

Henrik Plate

Michail Vasiltschenko

Michele Bezzi

Cedric R J Hebert

Alessandro Sorniotti

Anton Snitko

Elton Mathias

Stuart Short

Carlo Maria Negri

Caroline Banchereau

Karim Oueslati

Aymen Mouelhi

Micele Bezzi

Skander Ben Mahmoud

Sofiane Lounici

In an example embodiment, a solution is provided for an automated penetration testing system using pre-trained large language models (LLMs), which come equipped with a knowledge base of existing system configurations and commands. The knowledge base of these pre-trained LLMs may be extended with new training data, thereby updating their knowledge with new configuration commands, systems, and so forth.

Generative artificial intelligence penetration testing automation

In an example embodiment, an efficient, automated method to generate password guesses is provided by leveraging online text sources along with natural language processing techniques. Specifically, semantic structures in passwords are exploited to aid system in generating better guesses. This not only helps cover instances where traditional password meters would indicate a password is safe when it is not, but also makes the solution robust against fast-evolving domains such as new slang in natural languages or new vocabulary arising from new products, product updates, and services.

Machine-learning for password guess risk determination

In an example embodiment, a combination of machine learning and rule-based techniques are used to automatically detect social engineering attacks in a computer system. More particularly, three phases of detection are utilized on communications in a thread or stream of communications: attack contextualization, intention classification, and security policy violation detection. Each phase of detection causes a score to be generated that is reflective of the degree of danger in the thread or stream of communications, and these scores may then be combined into a single global social engineering attack score, which then may be used to determined appropriate actions to deal with the attack if it transgresses a threshold.

Computer system attack detection

Source code is scanned to generate a list of vulnerable tokens. Thereafter, the list of vulnerable tokens is inputted into a machine learning model to identify false positives in the list of vulnerable tokens. Based on this identification, the list of vulnerable tokens can be modified to remove the identified false positives. Related apparatus, systems, techniques and articles are also described.

Low false positive token identification in source code repositories using machine learning

Various examples are directed to systems and methods for identifying textual information regarding a first topic. A computer system may access a plurality of text units and detect that a first text unit of the plurality of text units is in a first language. The computer system may access a first language keyword set for the first topic, where the first language keyword set comprises a first plurality of keywords associated with the first language. The computer system may determine a first relevance score for the first text unit based at least in part on the first language keyword set. If the first relevance score is greater than a relevance score threshold, the computer system translates the first text unit to a base language and determines a text unit classification for the first text unit using a classification model trained with training data in the base language.

Computer systems for classifying multilingual text

A user authentication process is initiated by a software application executing on a primary server. The user authentication process prompts a user to enter, via a graphical user interface, login credentials. Thereafter, it is determined that the login credentials have been flagged. The user is then directed from the primary server to a fake server (i.e., a second server) mimicking the software application executing on the primary server. Thereafter, the fake server obtains metadata associated with the user interacting with the fake server that characterizes the user. This obtained metadata can later be provided (e.g., displayed, loaded into memory, stored in physical persistence, transmitted to a remote computing system, etc.). Related apparatus, systems, techniques and articles are also described.

Dissuading stolen password reuse

Embodiments provide systems and methods configured to mine information available from informal sources (e.g., social media, blogs, and forums) regarding security vulnerabilities. Particular embodiments may comprise engine(s) of a backend in communication with a user through an interface of a frontend, and also in communication with an underlying database to store security information and related information (e.g. search parameters). Embodiments may allow creation of user-specific search phrases for searching information in one or more informal social media information sources. Search results may be consolidated, and users such as system administrators quickly alerted to possible security issues. Embodiments may refine data mining over time by tracking the reputation (e.g. for data accuracy, freshness) of various sources. Embodiments may also reference formal official and third party sources of security information.

Mining security vulnerabilities available from social media

Embodiments automate tracking of exploit information related to initially-identified security vulnerabilities, through the data mining of social networks. Certain social network communities (e.g., those frequented by hackers) share information about computer security breaches (zero-day events). Embodiments recognize that further relevant security information may be revealed, in conjunction with and/or subsequent to such initial zero-day vulnerability disclosures. That additional information can include valuable details regarding known (or unknown) vulnerabilities, exploit codes and methodologies, patches, etc. Tracking that additional information can benefit security researchers/experts/law enforcement personnel. Embodiments monitoring social media traffic based upon initial security vulnerability information, perform analysis to detect patterns and create relevant keywords therefrom. Those keywords in turn form a basis for generating social media stream(s) responsible for harvesting additional security-relevant data. Results of further analysis of the social media stream can be fed back in an iterative manner to refine pattern detection, keyword creation, and media stream generation.

Automated security vulnerability exploit tracking on social media

Anonymity and confidentiality of information published from a microblogging platform, are preserved using randomly chosen relays (not related to the publisher account) in order to hide content in the cloud of published messages. The information can be relayed in clear text or in encrypted format. Additional linked relays may be used to overcome character number limitations imposed by the microblogging platform, with the longer full text of the original message reconstructed at the conclusion of the process. Depending upon the desired degree of confidentiality, complexity of the relay combination can be adjusted, and the path secretly shared among sender and authorized recipient. Only authorized recipient(s) can obtain (through another platform) the path combination to reach the message. A trusted third party stores the path relays and authorizations to access the path. The confidential information that is to be shared, may remain on the microblogging platform spread randomly over anonymous accounts.

Anonymous sharing of microblog publication

A report handler may receive abuse reports from reporters alleging policy violations of network use policies by at least one potential victim, and a source analyzer may determine at least one subset of the reporters. A content analyzer may determine a reference to the at least one potential victim in network activities of the at least one subset, and a review requester may generate a notification of a potential coalition attack against the at least one potential victim, based on the reference in the context of the at least one subset.

Prevention of coalition attacks in social network communities

A request handler may be configured to receive an enforcement request for enforcement of an obligation required as a condition for a previously-granted first resource access request. n obligation enforcer may be configured to enforce the obligation, based on the enforcement request, and a compliance manager may be configured to obtain certification of execution of the obligation from an obligation certification service, and to provide the certification as a basis for granting a second resource access request.

Obligation enforcement for resource access control

Provided personal information from a user may be determined, the provided personal information being associated with network publication thereof. A comparison of the provided personal information with password-related information may be performed. Based on the comparison, it may be determined that a risk level associated with the network publication relative to password security of at least one password associated with the password-related information exceeds a predetermined risk level. The user may be notified that the network publication of the provided personal information is associated with potential compromise of the password security of the at least one password.

Password protection using personal information

According to a general aspect, a computer-implemented method for a first user to verify an association with a second user through a secret handshake protocol includes maintaining information about a reusable identification handle for the first user, where the information about the reusable identification handle is provided by a trusted third party, maintaining information about a reusable credential for the first user, where the information about the reusable credential is provided by a trusted third party, and maintaining information about a matching reference for verifying an association with another user, where the information about the matching reference is provided by a trusted third party. Information based on the reusable identification handle and based on the reusable credential is transmitted to a potential peer. First information based on a reusable identification handle for the second user is received, and second information based on a reusable credential for the second user is received. A first comparison of a combination of the first information and the second information is performed with the matching reference to determine whether the second user's credentials match the first users matching reference. A second comparison of the first information with information published on a revocation list is performed to determine whether the second user's credentials have been revoked from usage. Based on the first comparison and the second comparison, a determination is made whether or not to verify the association of second user with the first user.

Growing community of inventors

Nice, France

Slim Trabelsi