Why do you think VAARHAFT made the finalists list and why is it Deep Tech?
We believe that VAARHAFT has made it onto the list of finalists due to its innovative solution and the relevance of its technology. VAARHAFT’s “Fraud Scanner” checks the credibility of digital images in seconds and reliably detects image-based fraud. The solution uses state-of-the-art AI technologies to detect AI-generated and manipulated images and offers unique features such as the localization of edited locations and cross-organizational duplicate checks.
VAARHAFT is a deep tech company because its technological core is based on complex computer vision technologies and deep neural networks (deep learning).
What is the key technological innovation within Fraud Scanner and why does it advance the state of the art?
The main technological innovation in the “Fraud Scanner” is the combination of complex computer vision techniques and deep learning mechanisms to recognize AI-generated and processed images. Compared to traditional methods, which rely primarily on metadata analysis and analytical approaches from image forensics, the Fraud Scanner offers faster and more reliable detection of deepfakes and image manipulation. This innovation advances the state of the art because our model achieves very high accuracy and we can respond to new image generators within 48 hours thanks to particularly fast development pipelines. Thanks to our proprietary meta-learning approach, the Fraud Scanner even recognizes images from previously unseen AI models. In addition, our solution can be flexibly integrated into existing processes as an API and meets all data protection requirements, as no images are stored or used for further training. This combination of high accuracy, fast adaptability and
flexible integration makes the Fraud Scanner a leading solution in the field of image-based fraud detection.
What was the biggest obstacle in your developing process? How did you overcome it?
Technologically, developing a classification model to recognize AI-generated and processed images was a complex task that required intensive research and development work. We overcame the challenge of continuously adapting to new GenAI developments by using advanced techniques such as meta-learning and optimizing our development pipelines to respond to new threats within 48 hours. Regular testing and pilot projects with real data allow us to continuously improve the models and increase their accuracy. Creating extensive and high-quality data sets is also a challenge.
However, here too we have now established good data mining structures that allow us to generate new, high-quality data sets very quickly. Accessing the German insurance market was also a major challenge. Our involvement in the established InsurTech networks “InsurLab Germany” and “New Players Network” of Versicherungsforen Leipzig proved to be particularly valuable in increasing our market presence and establishing valuable contacts in the insurance industry.
Another significant obstacle was ensuring data protection compliance, particularly with regard to the General Data Protection Regulation (GDPR). As our fraud scanner works with sensitive image data, it was essential to adhere to the highest data protection standards and develop our feature in compliance with data protection regulations.
Can you describe the unique innovation Fraud Scanner brings to e.g. the insurance sector, and how do you foresee it impacting both the markets, vulnerable to image manipulation, and society at large in the next five years?
Our AI technology in our Fraud Scanner provides the insurance industry with the unique innovation that fraud cases that were previously impossible to detect can now be prevented. The combination of a growing shortage of skilled workers and increasing automation in particular means that more and more claims settlement processes are being automated, making them particularly susceptible to image-based fraud. With our solution, we safeguard against increasing automation so that image-based insurance fraud can be countered despite increasing automation. This enables insurance companies to actively reduce their fraud rate and save process costs as well as fraud costs.
Over the next five years, our technology will have a significant impact on markets that are vulnerable to manipulated images and where the credibility of digital images is essential. As publicly available AIs become better at generating and manipulating digital images, the demand for reliable solutions to authenticate digital content will increase. Our technology will enable companies to address the growing challenges of deepfakes and other image manipulation by quickly and reliably detecting forgeries.
For society in general, this means a step towards greater security and trust in digital media. At a time when the credibility of digital images is rapidly declining, our technology offers a way to restore this credibility and ensure that digital content is reliable and public trust in digital media is restored.
How does Fraud Scanner integrate with existing systems, and what collaborative efforts have you engaged in to enhance its functionality and reach?
Our Fraud Scanner is available as a software interface (API) and can therefore be seamlessly integrated into existing systems. The API can be easily integrated into existing processes without replacing existing software, but rather only complementing or extending it in a meaningful way. In addition, we also offer a user interface for manual checks, where nothing needs to be integrated or downloaded, but users simply have to log in once with their login data.
To improve functionality and reach, we regularly present our solution at specialist conferences and events. There, we learn a lot about our customers’ needs and requirements in direct, personal discussions. We are also currently testing our solution in initial pilot projects with some insurance companies, where we are receiving valuable feedback and extensive insights that will help us move forward. We are also involved in networks such as InsurLab Germany and the New Players Network of Versicherungsforen Leipzig. This provides us with many important connections and collaborations in the insurance sector.
Experts and Society ongoingly discuss the ethical risks AI holds. Since you are the AI experts from the tech side, do we ensure technological innovation while avoiding ethical mistakes? Does Fraud Scanner respect ethical concerns, particularly in regarding bias, privacy, and transparency?
As AI experts, we are aware of the ethical risks and do everything we can to minimize them while still being innovative. And yes, this can be combined to a certain extent. When developing and training the Fraud Scanner itself, we paid close attention to bias, data protection and transparency.
To avoid bias, we train our AI models with diverse and representative data sets. Our meta-learning approach enables the Fraud Scanner to recognize even unknown AI models, which increases the fairness of the system. We make sure that our models do not exhibit any systematic biases.
Data protection is a key concern for us. For example, the Fraud Scanner was developed in such a way that it does not require any other (personal) data in addition to the image to be analyzed in order to assess the credibility of an image. The Fraud Scanner does not store any damage images and does not use them to train the models. Nevertheless, some images are sensitive data and must be extensively protected only during analysis. We are currently working on appropriate measures and strictly adhere to the General Data Protection Regulation (GDPR) in order to protect the rights and privacy of our customers.
Transparency is also important and is sometimes not so easy to enforce with AIs. Since an AI makes the decision about the credibility of the image, we can never explain 100% transparently why this decision was made. However, we offer our customers detailed explanations of the results of the fraud scanner. We highlight the detected areas of manipulation in the image in color and return the confidence values of the analysis. This makes it easier for our customers to understand the AI’s decisions. However, we are always transparent about the fact that our solution can only provide an indication and never constitutes court-proof evidence.
What are the next steps for VAARHAFT, and how do you plan to stay ahead in the rapidly evolving AI landscape?
The next steps for VAARHAFT are focused on further improving our technology and expanding our market presence. A key focus is on the continuous development of our fraud scanner to ensure it is always up-to-date in detecting image manipulation and AI-generated content. This includes constantly updating our detection models and building additional data sets. To stay at the forefront of the rapidly evolving AI landscape, we plan to further optimize our development pipelines to update our detection model even faster and enable high detection rates. We rely on advanced approaches such as meta-learning to recognize even unknown AI models. However, it is also important that our AI solution fits our customers’ needs perfectly. We attach great importance to feedback from pilot customers and partners in order to continuously improve our products and tailor them precisely to the needs of the market. We will be launching many exciting pilot projects with insurance companies in the near
future and hope to gain many valuable insights. This close cooperation will help us to optimize our solutions for practical use.
In addition to the German insurance market, where we already have a good network, we are also planning to expand into other European and international markets. Our aim is also to introduce our technology to other sectors. The fintech sector, for example, is very interesting for us. Here, however, we first need to undertake extensive market exploration to see how our solution can create the best added value.
Another important next step is ISO27001 certification. This is a really big chunk for us as a start-up, but it also gives us a huge boost in terms of data protection and data security. We are not only making our product GDPR-compliant, but also our processes within the company. With an innovative AI solution and a data protection-compliant way of using it, we are confident that VAARHAFT will maintain and expand its strong position in the AI landscape as a European provider.
What do you wish for in our technological or digital future?
We would like to see several key developments for our technological and digital future:
Firstly, it is very important to us that media literacy education is given a permanent place in school lessons. Pupils should learn early on how to critically scrutinize digital media and recognize when content has been manipulated, especially with regard to deepfakes and other forms of digital deception. This education and awareness-raising is essential to prepare the young generation for the challenges and potential dangers and to enable them to continue to form their opinions as freely as possible.
Secondly, we would like to see more “safe spaces” in which corporates and start-ups can work together quickly and easily. Such innovation spaces and industry-specific collaboration programs, such as InsurLab Germany or Versicherungsforen Leipzig, promote the exchange and joint development of new products by reducing bureaucratic hurdles and enabling creative collaboration. In our opinion, these environments are crucial for efficiently putting innovative ideas into practice and accelerating technological progress. Finally, it is important to us that the EU does not introduce further and even stricter restrictions in the field of artificial intelligence (AI). Instead, existing regulations should rather be reconsidered and, if necessary, adapted to ensure a balance between safety concerns and freedom of innovation. Excessive regulation could limit the potential of AI technology and set Europe back in global competition. It is crucial that we create a framework that
encourages both the development and application of AI without neglecting safety and ethical standards.
What advice would you like to pass to young entrepreneurs and tech innovators who want to reach their goals and maybe also want to take part in the Deep Tech Award 2025? Since out of 100 applicants you made it into the finalists, we figure you have some real pro-tips here.
We probably don’t need to give the advice that you should make mistakes early on, that’s part of being an entrepreneur or inventor. Our advice is: Never let yourself be guided or influenced by just one opinion. As a founder, you can listen to many opinions and advice, and many of them may be very valuable. But many are not, and not every tip will suit your startup and your idea. It takes a lot of sensitivity to filter out the “sensible” advice.
In the end, however, you will always be “alone” (or at least in a team) with your decision for your startup. Be brave enough to make these decisions “alone”. And very importantly. Talk to your potential users as early as possible to validate the problem and the solution. Go out into the market and talk to as many potential customers as possible to find out early on what is really needed and where there is and isn’t a problem. This is the only way you can perfectly align your solution with the needs of your customers and create a need that will carry you far enough afterwards.
Thank you so much for all your fascinating insights!
