What does a startup do when its technology is so new that even investors can't categorize it at first? The Berlin-based company brighter AI knows this situation well. As one of Europe's first generative AI startups, the team led by founder Marian Klee initially even had to remove the term "generative AI" from their pitch decks because the market wasn't ready for it yet. A few years later, brighter AI was named "Hottest European Startup" by Nvidia, gained Deutsche Bahn as an investor, and was acquired by the Canon Group in 2025. A key building block on this journey: four research projects implemented in 2023 that were officially recognized by the state as innovative. In 2024, this resulted in a research allowance totaling over one million euros, applied for in collaboration with re:fund.
Data protection or data usage? For brighter AI, that's the wrong question
The basic idea behind brighter AI sounds simple but is technologically highly complex. Traditional anonymization, such as pixelation, protects personal data but simultaneously destroys its value for AI training and analysis. "Pixelation is like a sensor error," explains Marian Klee. "Data protection is important, and AI is important. It shouldn't be the case that Europe has to choose between the two. So we said that's the wrong question. The right question is how we solve it."
brighter AI's answer: anonymization using synthetically generated faces and license plates. The data remains fully usable but no longer contains real identities. This exact unique selling point was later one of the reasons for the acquisition by the Canon Group.
Four research projects that were ahead of their time
The projects conducted and funded in 2023 covered the entire spectrum of this vision. Research included a method for using AI anonymization in distributed hyperscale computing systems, real-time anonymization even with challenging target faces, the automatic detection of personally identifiable information in complex data sources, and an edge system for decentralized anonymization directly at the network periphery.
Real-time anonymization, in particular, arose from a concrete market need. Customers didn't just want to anonymize individual images or videos after the fact; they wanted to use live cameras directly at the source in a data-compliant manner. "That was a very relevant project and ultimately led to us receiving an investment from Deutsche Bahn," says Klee.
From security camera to smart city
What sounds technical on paper quickly hit the streets. The example of Deutsche Bahn shows how the research has changed things in practice. The company operates about 11,000 cameras at its roughly 750 train stations alone, plus thousands more on trains. But before the collaboration, the railway didn't use a single one of these cameras for innovation; they were used exclusively for security. "Yet cameras are one of the richest sensors for understanding the environment and bringing smart cities to life. It's simply wasted potential if these cameras cannot be used," explains Klee. "Everyone talks about making cities and railways smarter. Companies like ours are what ultimately make that possible." With brighter AI's live anonymization, cameras could for the first time be used simultaneously for security and smart city innovation.
brighter AI also worked closely with the BVG. Initial pilots turned into permanent projects in buses and trams, and to this day, trams equipped with the anonymization technology are operating in Berlin.
At Volkswagen, factory data was anonymized. This wasn't just about the GDPR, but primarily about protecting employees, as works councils are particularly strict regarding data protection. Only through anonymization could the data be used for innovation in production. Meta, Intel, and Nvidia are also among their customers; brighter AI worked with Meta and 12 other universities on datasets for the development of smart glasses, among other things. "All four projects were fundamental to the company, to what we were able to do in the market, and to the value of the company," says Klee.
State recognition as proof of technological leadership
The research allowance paid out in 2024 for the projects was far more than just a financial boost for brighter AI. The impact on investors was particularly noticeable. "The research relevance of the technology is externally confirmed. We had it in black and white that we are cutting edge. That is a proof point for investors," says Klee. The state audit thus became a seal of quality that not only secured public funding but also made it easier for private investors to decide to continue investing.
The challenge: Anyone researching technologies that the market doesn't yet understand must first make this level of innovation understandable to the auditing authorities. "When you do something in the market that no one understands yet, but it's so innovative that it requires a research allowance, you have to be able to explain it. Having a bridge like re:fund for that is essential," says Klee.
From Berlin pilot project to global expansion
Today, the value of that early research is clear. Around 30 percent of brighter AI's revenue now comes from Japan, in addition to customers in the USA and Brazil. The technology that was researched and funded in Berlin is now being used worldwide. Since the acquisition by the Canon Group in 2025, brighter AI has continued its journey under the Milestone umbrella.
Looking back, Marian Klee assesses the role of the many people involved in this journey as follows:
"A startup's success is a collective effort. If everyone who supports us contributes ten percent, it all adds up to a hundred percent in the end. And re:fund definitely contributed more than ten percent."
.png)
.png)
