European startups are eagerly adopting DeepSeek’s AI technology despite significant safety and ethical concerns. The model’s affordability has made it an attractive alternative, but experts advise caution.
'The goal of 🇺🇸 export controls has always been to cripple 🇨🇳’s AI sector. But, as DeepSeek has shown, they have had the opposite effect, spurring precisely the innovations that will enable 🇨🇳 firms to challenge 🇺🇸 AI oligopolies,' writes @AngelaZhangHK.https://t.co/2z8oNUfwIL
— ASPI (@ASPI_org) February 6, 2025
DeepSeek, a Chinese AI developer, stunned the tech world last month by rolling out its high-performing large language model (LLM), R1, for less than $6 million.
This cost is significantly lower compared to similar developments by industry giants such as Google and OpenAI. Just two weeks after its release, several European startups have started integrating R1 into their products. Elevenlabs, a synthetic voice startup, was among the first to announce the integration of DeepSeek’s R1 model.
This @TechRepublic article mentions Andrew Horton and @CoreraJames's recent article in The Strategist on DeepSeek's R1 AI model breakthrough and the need for a comprehensive Australian tech strategy that covers AI.
Read it now 👉 https://t.co/5TD9tBRW5Uhttps://t.co/UYw7MrdLOl
— ASPI (@ASPI_org) February 5, 2025
London-based AI unicorn Synthesia is also experimenting with the model to evaluate its performance.
DeepSeek is a significant advancement for AI and open-source. But it's essential to remember that an AI model cannot be a censored propaganda machine for China or, even worse, propagating falsehood. pic.twitter.com/vPqce4JE4H
— Aravind Srinivas (@AravSrinivas) February 4, 2025
Denis Kalinin, head of APAC at Runa Capital, mentions that numerous German and European startups are exploring the integration of DeepSeek’s technology. However, deploying DeepSeek’s models is not without risks.
Critics have raised concerns over the model’s apparent biases, noting that it often refuses to generate content critical of the Chinese government. Privacy and security experts warn that improper use of this technology could lead to legal repercussions. Dominik Lambersy, co-founder of Berlin-based AI startup TextCortex, emphasized the cost benefits after testing DeepSeek’s integration.
“The cost is ridiculous. It’s so cheap in comparison to everything else,” Lambersy stated.
Startups weigh cost of DeepSeek
The company is considering making DeepSeek’s model the default for its new users. German AI companies Langdock and Novo AI have confirmed deploying DeepSeek’s models in their systems. Swedish notetaking startup Sana AI also made a similar announcement.
Several other startups are experimenting with the model to gauge its capabilities. Startups like Synthesia and Tessl are running tests to determine if DeepSeek’s model meets their standards for performance, security, and integration. Jose Gaytan de Ayala, an investor at Kinnevik, noted that many AI coding companies are contemplating integrating DeepSeek due to its favorable cost-performance ratio.
Despite its popularity, there are significant data privacy concerns regarding DeepSeek’s API usage directly from the Chinese cloud. Although the model can be hosted on local servers in Europe to mitigate some risks, regulatory scrutiny remains intense. Italy has already banned the use of DeepSeek’s model, and regulators in France, Belgium, Taiwan, South Korea, and the US are investigating the platform.
This regulatory environment has made some sectors, especially those handling sensitive data like legal tech, HR tech, and fintech, more cautious about deploying DeepSeek’s models. A study by LatticeFlow, a Swiss trustworthy AI company, suggested that DeepSeek’s models might not comply with the EU’s forthcoming AI Act, highlighting security vulnerabilities and higher bias levels compared to competitors. Peter Tsankov, CEO at LatticeFlow, emphasized the ongoing issues, “While progress has been made in improving capabilities and reducing costs, there are critical gaps in cybersecurity, bias, and censorship.”
Despite these concerns, some industry experts believe the initial worries will diminish over time.
Rasmus Rothe, General Partner at Merantix Capital, commented, “You can deploy it safely, but there are still risks of the model providing responses with clear biases. However, the focus on cost efficiency will likely dominate the industry’s attention in the near future.”
As startups and regulators navigate the benefits and risks of integrating DeepSeek’s AI models, the debate continues over the balance between innovation and safety in AI deployment.