AI Security & Testing
Artificial intelligence not only brings with it new possibilities, but also new attack surfaces. Unlike traditional software, AI systems behave dynamically and are not entirely predictable. This makes them vulnerable to targeted manipulation, unwanted model distortions, and security-relevant weaknesses in the database or system integration.
Our particular strength lies in the combination of classic vulnerability management with modern AI testing.
As an official partner of the global market leaders in this field, we have proven tools and in-depth expertise to systematically analyze even complex AI systems for vulnerabilities – from the training data set to the interface to the production environment.
Our expertise is based not only on practical experience but also on our active participation in shaping standards. For example, we were directly involved in the creation of a guideline for penetration testing of large language models, which was developed in collaboration with the Federal Office for Information Security (BSI) (publication in preparation).
We simulate real-world attack scenarios on machine learning models, test their robustness against adversarial inputs, and evaluate potential risks such as model theft, data poisoning, or unwanted behavioral changes due to model drift. We think not only in terms of algorithms, but also in terms of entire systems.
Our customers benefit from a security architecture that doesn't treat AI as a special case, but consistently integrates it into existing security and compliance structures. Whether internally developed models or third-party AI – we help bring trust and transparency to learning systems.