AI testing and monitoring specialist Advai will help shape the design, evaluation and governance of Cosine's planned sovereign model
London, UK — Cosine today announced that Advai has joined its sovereign AI coalition, bringing independent AI testing, evaluation and monitoring expertise to the development of sovereign AI capability for regulated and high-assurance environments.
Advai will help direct the design and evaluation priorities of Cosine's planned Lumen Sovereign model. Its contribution will focus on ensuring that assurance and safety requirements inform model development from the outset, rather than being assessed only after capability work is complete.
The collaboration will translate practical assurance needs into model-design decisions, evaluation criteria and measurable thresholds. Advai will contribute expertise across risk identification, threat modelling, adversarial testing, safety, security, robustness, auditability and continuous monitoring.
"Sovereign AI goes beyond where a model runs or the data is stored. Organisations also need evidence that it behaves as intended, that its failure modes are understood and that it can be governed over time," said Alistair Pullen, CEO of Cosine. "Advai brings the assurance expertise needed to shape those requirements early. Their contribution will help us make safety, security and accountability a core part of the model's design and development process."
Advai’s platform provides independent testing, evaluation and assurance across the AI lifecycle, helping organisations select the right models, assess their readiness for deployment and monitor how they perform in the real world. It combines dynamic benchmarking and adversarial testing with traceable evaluation frameworks and continuous monitoring, enabling organisations to identify emerging risks, detect performance or security issues and maintain confidence in AI systems as they evolve.
“We are excited to be collaborating with Cosine on the Lumen Sovereign model build, bringing Advai’s extensive experience in testing, red teaming and monitoring AI models and systems to the coalition”, said David Sully, CEO of Advai. “If sovereign AI is going to be trusted in high-assurance and regulated environments, safety, security and accountability need to shape how those models are designed and built from the ground-up. Doing so creates a foundation for effective, reliable and trustworthy models that can be deployed at scale.”
As part of the coalition, Advai will help shape an assurance-led approach including:
● Defining safety, security, robustness and reliability requirements.
● Designing evaluations for normal, adversarial and operationally realistic conditions.
● Establishing evidence and thresholds to support development and deployment decisions.
● Identifying monitoring signals for drift, regressions and unsafe behaviour.
● Connecting technical evaluation with governance, accountability and audit requirements.
Cosine is developing Lumen Sovereign as a planned UK sovereign model for software engineering and other high-value tool-using work. The programme is focused on controlled model development, deployment independence, reviewable workflows, evaluation and auditability for environments where foreign-hosted AI may be unusable.
The coalition brings together organisations with complementary expertise to ensure sovereign AI is shaped around real operational requirements. Advai's participation adds an independent assurance and safety perspective to Cosine's work on post-training, model evaluation and controlled deployment.
About Cosine
Cosine is a British AI lab building Lumen Sovereign for defence, national security and other tightly regulated environments. We train and post-train models for software engineering and other tool-using work that has to run inside high-security systems. The programme is supported by the UK Government’s Sovereign AI initiative and is being developed on UK sovereign compute with UK-controlled training and governance. Customers can run Cosine inside their own VPC or fully on-premises, with no data egress and much tighter control over deployment, data handling and infrastructure. For more information, visit cosine.ai.
About Advai
Advai is an independent AI assurance company helping organisations adopt and deploy AI with confidence. Its platform delivers fast, scalable and context-aware testing across AI systems, evaluating performance, security, robustness and bias in real-world conditions.
Combining deep technical testing with expert human judgement, Advai provides objective insights across the AI lifecycle, from model selection and procurement through to deployment and continuous monitoring.
Headquartered in the UK with global reach, Advai works with leading organisations across finance, defence, insurance and the public sector, helping them deploy AI safely, securely and responsibly.
