Cloudflare announced the addition of the open AI models EuroLLM and Apertus to the Workers AI platform, alongside the launch of practical workshops to help government entities and critical infrastructure operators build cyber defenses that do not depend on a single model. The move comes a year after the company introduced the concept of “choice” as a foundation of sovereignty in AI.
Open European Models on Workers AI
EuroLLM supports a total of 35 languages, including the European Union’s 24 official languages, and was developed with support from Horizon Europe, the European Research Council, and EuroHPC through a consortium comprising European universities and research institutions, Unbabel, and Naver Labs. The model was trained on the MareNostrum 5 supercomputer, and the consortium says it outperforms similarly sized models in multilingual and European Union-specific machine translation tests. Access to it can be requested through Workers AI.
Apertus, whose name means “open,” is Switzerland’s first fully open, large-scale multilingual language model. It was trained on more than 15 trillion tokens across more than 1,500 languages, with 40% of its training data allocated to non-English languages. It was developed by ETH Zurich, EPFL, and the Swiss National Supercomputing Centre as part of the Swiss AI Initiative, and its architecture, weights, training data, and methods were published. It was also designed with Swiss and European rules in mind, including the European AI Act and GDPR, including respect for data-exclusion requests, the removal of personal data, and limits on retaining it verbatim.
What Changes in Practice?
The addition gives governments and developers the option of using models linked to local linguistic and regulatory contexts, rather than limiting applications to closed commercial models. Cloudflare points to examples that it says developers, students, and government employees created over the past year, including Form Mitra in India, which guides users by voice through eligibility forms in 22 Indian languages; MedBridge in Singapore, which supports health conversations in 14 languages; and Anshin Concierge in Japan, which connects a user’s description of their problem with the appropriate government service.
Defenses That Do Not Stop at a Single Provider
Cloudflare says its security team developed an orchestration layer that combines multiple models operating in parallel to search for vulnerabilities, verify results, and rank threats. The company has released this layer as open source, so it can be run with closed or open models selected by the user. The core idea is that losing access to a single provider does not automatically disable the defense system.
The company is launching hands-on workshops for government cybersecurity agencies and critical infrastructure operators, during which participants will learn to build the orchestration tool and layered defenses and adapt them to their environments and rules. The first workshop will be held in Singapore during October as part of Singapore International Cyber Week.
certi.news Analysis
The important shift here is not merely adding two models to a cloud platform, but combining two open multilingual options with security tools that are not tied to a specific provider. This matters to entities that consider continuity of access and linguistic and regulatory compliance part of national resilience. However, the source does not provide independent results measuring the performance of the two models or operational details about the workshop requirements and supported environments; therefore, claims about superiority and practical impact remain based primarily on what Cloudflare and the developing organizations reported.