Cosine unites UK industry to build Britain’s first sovereign frontier AI model

British AI lab Cosine has secured support from a group of major UK companies to co-develop what it calls the country’s first sovereign frontier artificial intelligence model, Lumen Sovereign. The project is positioned at the intersection of national security, financial stability and technological independence, as large institutions seek to reduce reliance on foreign AI providers.

Backed by the UK government’s Sovereign AI programme, the initiative aims to create a high-end model that can be deployed entirely within domestic infrastructure, with a particular focus on defence, national security and tightly regulated sectors such as banking and healthcare.

Coalition of leading UK institutions

Cosine, founded in London in 2023, has brought together a coalition of prominent organisations to shape the design of Lumen Sovereign. Companies that have signed memoranda of understanding include Babcock International Group, BT, Lloyds Banking Group, London Stock Exchange Group (LSEG), NatWest Group, PwC, Thales UK, BAE Systems, Leonardo UK and Telefónica Tech UK&I.

These partners will collaborate with Cosine during the model’s design and development phase. Their role will be to help define concrete use cases, technical and security requirements, and governance standards that the model must satisfy before it is rolled out to production environments.

Strategic dependency and AI sovereignty

The formation of the coalition reflects growing concern among large UK institutions that AI is becoming a critical dependency similar to energy supply or communications infrastructure. Many organisations in defence, finance and the public sector are reassessing the risks of depending on foreign-controlled AI platforms for core operations and sensitive data.

For some, the ability to control where models are trained, where they run, and how data is handled is moving from a technical preference to a strategic requirement. Cosine’s approach is to offer a model that can operate entirely within a customer’s own infrastructure, including air-gapped systems, without external data transfer or reliance on overseas platforms.

Training on Isambard-AI under the UK Sovereign AI programme

Lumen Sovereign will be trained exclusively on Isambard-AI, described as one of Europe’s most powerful supercomputers. Access to this compute has been awarded through the UK government’s £500 million Sovereign AI programme, which is intended to strengthen the country’s domestic AI capabilities.

Cosine emphasises that the model will be built from the ground up using proprietary datasets. Rather than fine-tuning an existing open-source checkpoint, the company plans to use in-house data across all stages of training, from pre-training through to post-training optimisation.

Focus on regulated and high-sensitivity workflows

Cosine positions itself as a sovereign AI frontier lab, developing advanced models and coding agents designed specifically for defence, national security and highly regulated industries. According to the company, its platform is engineered so that the full lifecycle of the model – development, deployment and operation – can take place under UK jurisdiction and within a customer’s own infrastructure.

Lumen Sovereign is being tailored to support complex, sensitive workflows where traditional AI tools have struggled to gain acceptance due to regulatory, privacy or security constraints. Targeted early applications include:

• Cybersecurity and adversarial testing
• Know-your-customer (KYC) and anti-money laundering (AML) alert investigation
• Clinical trial coordination
• Legal document review
• Healthcare administration and related operational tasks

To address these use cases, Cosine has assembled what it describes as one of the largest collections of domain-specific training datasets outside of the hyperscale cloud providers. These datasets span more than 30 workflows across regulated industries, with the aim of making the model applicable across the full range of needs represented by coalition members.

Concerns over vendor lock-in and cost

Cosine’s leadership argues that reliance on a small number of foreign AI vendors creates several forms of risk for large enterprises: security exposure, strategic dependency and potentially rising long-term costs. By building a model trained entirely within the UK and capable of running in tightly controlled environments, the company is seeking to offer an alternative to large US-based AI platforms.

The firm also contends that its approach can deliver competitive performance at a lower cost than leading frontier models from providers such as OpenAI and Anthropic, particularly for organisations that need private deployments or air-gapped installations.

Industry, government and technology collaboration

BT’s innovation leadership has framed the project as an example of how telecoms, industry and government can work together to establish trusted digital infrastructure for the UK economy. As AI becomes embedded in core services, from banking systems to communications and public services, participating organisations are seeking stronger assurances about data handling, security and alignment with sector-specific regulations.

Lumen Sovereign is currently scheduled to reach deployment readiness by the end of 2026. If successful, the project could provide UK institutions in finance, defence and healthcare with a domestically controlled alternative to foreign frontier models, while potentially setting a template for how other countries approach AI sovereignty in critical sectors.