One year later: Sovereign AI and the fight for choice
Cloudflare expands AI model access with EuroLLM and Apertus while launching hands-on AI defense workshops for governments, emphasizing choice and sovereignty amid rising access restrictions.
Cloudflare marked its Birthday Week by releasing two new AI models: EuroLLM, covering all 24 official EU languages, and Apertus, Switzerland’s fully open multilingual model trained on over 1,500 languages. Both were developed by public universities and research institutions and are now available on Workers AI, with access requests open to the public. The company also introduced hands-on workshops to help government cyber agencies and critical infrastructure operators build AI defenses compatible with any model, starting with a session in Singapore in October.
The announcements follow a year of growing concerns about AI access and sovereignty, with some countries restricting access to frontier models based on geography. Cloudflare argues that choice—not isolation—is the solution, pointing to initiatives in India, Japan, and Singapore where open models enabled tools for rural citizens, elderly patients, and healthcare workers. The company’s security team developed an AI defense harness that coordinates multiple models to hunt vulnerabilities, verifying findings and prioritizing threats without dependency on a single provider.
EuroLLM, developed with support from Horizon Europe and EuroHPC, supports 35 languages, including all 24 official EU languages, and was trained on the MareNostrum 5 supercomputer. According to the consortium, it outperforms similar-sized models on EU multilingual benchmarks and machine translation tasks. Apertus, Switzerland’s first fully open multilingual model, was trained on over 15 trillion tokens across 1,500 languages, with 40% of data in non-English languages, and was designed to comply with EU AI Act and GDPR standards.
Cloudflare’s new AI defense workshops aim to help governments and critical infrastructure operators implement layered security architectures and adapt AI defenses to their environments. The first workshop will take place at Singapore International Cyber Week in October, with modules designed to integrate into national AI skilling and cyber resilience programs. Partners like CyberPeace in India and Code for Japan have already helped deploy open models into practical tools, reflecting a broader shift toward accessible, sovereign AI solutions.