Two phase 1 projects were evaluated positively:
The INPUTLAB project aims to improve the security of software applications that use structured data formats such as XML or JSON. Because these are often tested manually and inadequately, there is a risk that vulnerabilities will be exploited in attacks. INPUTLAB addresses this problem by automatically generating test data for these formats so that developers can thoroughly test their applications for errors. At launch INPUTLAB will provide test data for over 100 popular software applications, focusing on critical business areas such as finance, telecommunications and public administration.
>> https://cispa.de/inputlab
The primary goal of the future startup Trustlens' "Aletheia" project is to provide end users with a robust platform that enables them to check digital content for AI-generated deepfakes and efficiently extract and present useful information. It can serve as a tool for law enforcement, forensic experts, journalists, media service providers, data scientists and analysts in various industries to support data-driven decision making and detection processes. The aim of the work planned under the StartUpSecure Phase 1 funding is to extend the described platform with additional, innovative AI models for detection.
>> https://cispa.de/trustlens
Three additional projects have green light for their applications in phase 2:
Cyberfame will receive funding from the BMBF for the implementation of a project focusing on the development of an "Artificial General Cybersecurity Intelligence (AGCI)". It aims to comprehensively analyze and secure the open source ecosystem in order to protect Europe's digital infrastructure. Cyberfame's tool will be used to make software in the supply chain more secure. It enables the identification of vulnerabilities in individual open source repositories of complex software projects. The goal is to avoid exploits like Log4J in the future.
Under the name "CyberLinguaGuard", contexxt.ai is developing a solution that ensures the data security of knowledge in large, shared technologies. The focus is on two technologies: semantic databases and large language models. Semantic databases or knowledge graphs efficiently store knowledge and information for knowledge extraction, while language models formulate the extracted knowledge in a target-group-specific way, in this case with generic instead of personalized knowledge. These technologies serve as central knowledge assistants to efficiently preserve and communicate corporate knowledge with the help of innovative management tools. The challenge is to develop a solution that is trustworthy, secure and GDPR compliant.
PriMFlow is working with BMBF funding to develop mechanisms that allow users to revoke the storage of their customer data. The product innovation is the selection and adaptation of ML-Ops components that meet security and privacy requirements and enable the deletion of personal data. Technically, the team is developing missing interfaces and components in the ML-Ops architecture. The goal of the project is a demonstrator that provides a stable technological basis for a marketable product. It includes a configured ML flow system, security policies, implementations for deletion operations, risk stratification and data partitioning, as well as theoretical results and implementations for model manipulation.
