Startups and Entrepreneurship
CISPA seeks to have a strong and decisive global impact on society and economy by combining cutting-edge, often disruptive, foundational research with transfer to innovative applications.
An interdisciplinary team therefore supports prospective founders in the idea and start-up phase. This support includes the conception and realization of new business models in the field of IT security:
In our CISPA Startup Blog we regularly publish news, interviews and interesting facts.
Click here for an overview of all previous articles
The following startups are currently in incubation at CISPA. They receive professional support in product development, business development as well as marketing and funding.
Click here for an overview of all startups
Technology
Contexxt.ai leverages modern Large Language Models (LLMs) to optimize knowledge management for clients. Based in Saarland, Germany, with additional presence in Hamburg and Bavaria, the team prioritizes maximizing LLM benefits while upholding top-notch security and privacy. Their commitment goes beyond interest; they utilize LLMs as invaluable assets for clients through innovative software solutions. Collaborating closely with customers, they merge their ideas with cutting-edge technologies, delivering bespoke solutions tailored to unique business needs. This proactive approach to security enables businesses to optimize operations using advanced LLMs all while safeguarding the privacy of their data.
Wissenschaftliche Grundlage
Durch die Zusammenarbeit mit renommierten Forschungseinrichtungen wie dem CISPA Helmholtz-Zentrum zeigt contexxt.ai sein Engagement für bahnbrechende Innovationen, bei denen Sicherheit und Datenschutz in jeden Aspekt des Ansatzes integriert sind. Sicherheit und Datenschutz sind grundlegende Säulen der Strategie und tief in die empfohlenen Architekturen und Methoden integriert. Das Team erforscht aktiv die Sicherheit von Vektordatenbanken und erkennt deren Potenzial als optimale primäre Datenquellen für Sprachmodelle, insbesondere im Hinblick auf Datensicherheitsverletzungen.
In der aktuellen Forschung untersucht contexxt.ai, wie Vektordatenbanken die Mietertrennung durch die Integration von Wissenselementen mit eindeutigen TenantIDs erleichtern können. Dies beinhaltet die Integration von TenantIDs in die Vektordatenbank, die Implementierung von Datenpartitionierungstechniken und die Anwendung von mieterspezifischen Filtern. Diese Bemühungen zielen darauf ab, den Datenschutz und die Sicherheit in gemeinsam genutzten Sprachmodellen zu verbessern, was zu Vorteilen wie erhöhter Datenvertraulichkeit, Einhaltung gesetzlicher Vorschriften und gesteigertem Kundenvertrauen führt. Indem contexxt.ai der Sicherheit Priorität einräumt, unterstreicht das Unternehmen sein Engagement für den sorgsamen Umgang mit Daten und die Aufrechterhaltung des Vertrauens.
Management
Website
Technology
The weakness sequire technology discovered in 2023 in large language models (LLMs) is in the way this technology works. This means that it cannot be
simply "turned off". There is no simple, monocausal approach to fixing the problem. Instead, a mix of approaches to a homogeneous, secure environment for language models, i.e. an LLM operating system. In addition to the results of their own research, they use traditional elements for protection, such as process separation, rights management, hypervisors, etc, hypervisor, etc.
Scientific Background
sequire technology played a leading role in the discovery of the most important vulnerability in the area of Large Language Models (LLM).
In the current OWASP Top 10 for LLMs, the Indirect Prompt Injection vulnerability discovered by sequire technology is the number one threat;
The German Federal Office for Information Security (BSI) also warned about this in a separate publication.
The research combines own findings on LLM vulnerabilities with known principles from the field of operating systems and computer security.
systems and computer security to create a secure environment for running LLMs.
Management
Website
Technology
Simplyblock is a clustered cloud storage solution that provides users with the ability to create virtual (logical) block storage devices, that scale arbitrarily in size and speed, yet work just as simple as a locally attached disk. Built upon the industry-standard, and fast NVMe over Fabrics protocol, simplyblock combines a multitude of backing storage technologies (local and remote block storage, as well as object storage) but provides a single, holistic view of the logical device. Furthermore, the solution enables automatic tiered storage (moving data between backend storages for price and performance reasons), as well as industry-expected features, such as compression, deduplication, encryption, and more.
In addition to that, features like immediate snapshots, remote backups, automatic self-healing, and our intelligent data balancing algorithms, as well as enhanced disaster recovery mechanisms help mitigate the dangers of ransomware attacks, or other types of security breaches that cause data loss, data encryption, or data corruption. Simplyblock integrates advanced cybersecurity features to recognize and stop new attack vectors and offer an immediate recovery in situations where data was already modified, aiming to bring a RPO (Recovery-Point-Objective) of 0, hence no data loss, to the world.
We want to democratize the ability to use cloud storage the way the user needs it, not the way it is offered by cloud providers.
Scientific Background
The simplyblock team combines decades of experience in different field, product-related fields, such as distributed systems, storage solutions, cloud infrastructures. Our technology is based on industry-proven components like SPDK (the Storage Performance Development Kit), a framework supported by industry giants such as Intel, and uses NVMe over TCP as the underlying transport protocol, providing an out of the box experience available in Linux and Windows. The combination of components provides an easy to use and high-performance solution to companies running IO-intensive workloads, such as databases, in the cloud.
Management
Website
Technology
Hyde is a platform for consumers to monetize their data in a privacy-preserving manner, and for businesses to understand their customers better in a compliant way. Built on a state-of-the-art privacy foundation, Hyde provides transparency, control and data sovereignty for consumers, while allowing businesses to expand their customer data beyond their own premises. The vision of Hyde is to equip consumers with data agency and to equalize business data moats.
Scientific Background
Hyde builds on recent progress in Trusted Execution Environments (TEE). This technology allows to compute on data without exposing it to third parties. Beyond that, Hyde is heavily using Machine Learning Embedding technology to convey consumer taste like music, movies, or shopping. Hyde is led by Dr. Uwe Stoll with a PhD in Semantic Web and Machine Learning, and a long track record in applied AI. CTO Kyohei Hamaguchi (JP) has about ten years of experience in designing complex software systems with an emphasis on information security.
Management
Website
Technology
Kertos is a no-code SaaS solution connecting an organization’s entire infrastructure to manage personal data and fully automate privacy processes. By executing workflows, handling data subject requests, and building self-maintaining records of processing activities, Kertos makes manual privacy (GDPR) compliance a thing of the past. The integration of external systems plays a central role in the automation of data privacy tasks. This requires corresponding API keys, access tokens or comparable access information. Kertos develops a hybrid zero-trust architecture to make sure data transfer takes place only between the (potentially on-premise) worker nodes and the integrated systems. Thus, the Kertos backend itself never comes into contact with costumers’ API keys or customer data.
Scientific Background
Kertos focuses on the automation of processes, the secure management of privacy requirements and the simplification of everyday workflows for all stakeholders involved. The founders Dr. Kilian Schmidt, Johannes Hussak and Alexander Prams combine relevant practical experience and professional knowledge, which they acquired at renowned universities such as the Technical University of Munich and the Humboldt University of Berlin. While Dr. Kilian Schmidt, with his background in law and first-hand experience with legal processes, takes on the role of "legal expert" in the founding team, Johannes Hussak contributes expertise in product development and innovation as CPO and COO. The team is completed by Alexander Prams, who is responsible for the technical implementation of the Kertos solution in his role as CTO and with his field of expertise, automation.
Management
Website
Technology
LUBIS EDA has developed a software tool for the verification of digital integrated circuits (IC) which are part of a lot of semiconductors. With this tool, LUBIS EDA can automatically generate the Verification Intellectual Property (VIP). This VIP can then be used to ensure that there are no functional flaws and bugs within the IC. Functional bugs can not only lead to a malfunctioning semiconductor but can also be an entry point for security issues. The verification methodology is based on the so called “Formal Verification” technique for semiconductors.
Scientific background
Dr. Tobias Ludwig has developed a new and efficient methodology for the generation of the VIP within his time as a researcher at the TU Kaiserslautern. His dissertation resulted in a software prototype that implements this methodology. Dr. Michael Schwarz has done his dissertation at the same chair as Dr. Tobias Ludwig and is an expert in Formal Verification as well as Hardware-Software-Interfaces. Dr. Max Birtel holds a degree in business engineering and was a researcher at the German Research Centre for Artificial Intelligence (DFKI) and the SmartFactory Kaiserslautern in the field of Industrie 4.0 .
Management
Website
Technology
Natif.ai is a deep-tech startup in the field of intelligent document processing. The platform provided by natif.ai enables developers to automatically extract all relevant information from common documents via an API and pass it to downstream systems. In addition, developers can automate their own document processes without deep knowledge of modern AI techniques: the Natif platform guides developers through the entire setup process, trains use-case specific models, and makes them available via an easy-to-use API. The AI learns continuously through an active learning approach. Since all AI models are developed in-house and hosted in Germany (including an own deep-OCR), documents can not only be analyzed extremely fast and accurately, but also meet the highest data protection standards. Currently, natif.ai is working on AI-based forgery detection to assess whether documents have been tampered with. Thus, in the future natif.ai will support companies in both automating and securing document-based processes.
Scientific background
Natif.ai employs a large number of AI experts who studied at renowned universities and then either worked at research institutes such as the German Research Center for Artificial Intelligence (DFKI), some of whom earned their doctorates, or gained experience in the research departments of various commercial enterprises. Today, Natif.ai still works closely with the DFKI and CISPA, participates in funded research projects, and supervises bachelor's and master's theses.
Management
Website
Technology
NetBird is a next generation solution of network access and security that has been developed as open-source software since 2021. The platform combines Zero Trust principles with a highly scalable peer-to-peer network enabling organizations and teams to securely connect remote resources. It is based on Wireguard® and provides a fast and secure network for any use case, that verifies policies and client context at the edge, not at centralized gateways.
Scientific Background
NetBird has a team with over twenty years of experience in software and infrastructure engineering fields. They come from their countries' best information systems and computer networking universities with a focus on information security. They built their careers in data, automation, and security engineering, always providing the best user experience while keeping security a default option.
Management
Website
Technology
fuse.space is the first secure data and collaboration space that records and protects the complex and collaborative process of creative work, to protect its intellectual property. Our vision is to connect the world in creativity. A world where all people and entities can work together easily and securely, without having to know or trust each other. To do this, we are creating a space that supports, transparently documents, immutably secures, and verifiably protects the collaborative process of creative work. We address music creators, architects or scientists, anyone who shares and collaborates on their sensitive and unprotected intangible assets, such as ideas, concepts, and inventions. With the support of CISPA, fuse.space is further developing its solution to enable a secure collaboration space.
Scientific Background
With the help of over 1,000 creators and institutions in the creative industries, we are working together to create a solution that covers and secures the most important work processes and integrates them into their workflows. We ourselves come from the music industry and know the problem and its potential very well. Andre Angkasa has been building digital media services, products, and brands for over 18 years. Alexander Wittkowski, as a music producer and composer, knows the problem from his own experience. He had two of his songs stolen and published under a different name by a very famous pop star. Unfortunately, he lacks any proof of the collaboration process and his IP.
Management
Website
Technology
At the heart of the Xpect approach to modeling safe processes is a representation formalism that not only maps temporal sequences of actions, but also captures their semantics by linking them to existing knowledge. In this way, it becomes possible to prove central properties of abstract processes and concrete sequences and to correct them if necessary. By monitoring the current execution of a process, the violation of central security or compliance rules can be detected in advance and the occurrence of a critical state can be avoided.
Scientific Background
Most of Xpect's employees have many years of experience in artificial intelligence research. For example, many of them were employees at DFKI, with which there is still close cooperation. In addition, they have gained relevant experience in management consultancies or companies in large-scale industry, which pays off in particular when it comes to identifying relevant issues and assessing the weaknesses of current solutions.
Management
Website
Technology
The Cybervize platform is designed for organizations of all sizes to strengthen their cybersecurity. This solution includes an AI-based SaaS component that identifies risks, recommends cybersecurity measures, and assists with implementation. As an integral part of Cybervize, a moderated user forum promotes knowledge sharing and provides practical guidance. Human advisors provide targeted expert assessments to assist with implementation and conduct periodic reviews upon request.
Scientific Background
Cybervize's technology is based on the application of AI technologies and SaaS solutions to automate cybersecurity processes. The solution draws on proven cybersecurity practices and processes, as well as information security and risk management best practices, to provide a comprehensive and cost-effective cybersecurity solution specifically for small and medium-sized businesses.
Management
Website
CISPA supports a multitude of early stage venture teams, which are looking for co-founders and contributors. Applications for the following open positions will be evaluated continuously.