One year later: Sovereign AI and the fight for choice

Cloudflare announced during its Birthday Week that EuroLLM and Apertus, two open language models built by European public universities and research institutions, are coming to its Workers AI platform, with access requests open now. The company also said it is launching a program of hands-on workshops for government cybersecurity agencies and critical infrastructure operators, with the first running in Singapore this October at Singapore International Cyber Week.

EuroLLM supports 35 languages, including all 24 official EU languages, many of which the company says are underserved by existing open models. It was developed with support from Horizon Europe, the European Research Council and EuroHPC by a consortium including Instituto Superior Técnico, the University of Edinburgh, Instituto de Telecomunicações, Université Paris-Saclay, Unbabel, Sorbonne University, Naver Labs and the University of Amsterdam. It was trained on the MareNostrum 5 supercomputer and, according to the consortium, outperforms similar-sized models on EU multilingual benchmarks and machine translation.

Apertus, Latin for open, is described as Switzerland's first large-scale, fully open multilingual model. It was trained on more than 15 trillion tokens across more than 1,500 languages, with 40% of training data in languages other than English, by ETH Zurich, EPFL and the Swiss National Supercomputing Centre as part of the Swiss AI Initiative. Its architecture, weights, training data and methods are all published. It was designed with rules such as the EU AI Act and GDPR in mind, meaning respecting training opt-outs, removing personal data and preventing memorization. Training ran on CSCS's Alps supercomputer with more than 10,000 GH200 GPUs, and its developers report it significantly outperforms leading closed and open models on rare and regional languages, from Romansh and Swiss German to low-resource languages across Asia and Africa.

Cloudflare also described its own security work over the past year: a harness, or orchestration layer, that coordinates multiple AI models working in parallel to hunt for vulnerabilities, verify findings and prioritize threats. The company open-sourced the harness so organizations can run it with the models of their choice. The workshop program teaches participants to build their own AI security harness and layered defenses, with plug-and-play modules designed to slot into national AI skilling and cyber resilience programs.

Why it matters: Developers and public-sector teams gain another route to multilingual open models through Workers AI, and because the security harness works with any model, closed or open, losing access to one provider does not switch those defenses off.