Britain to build its first home-grown frontier AI – backed by £500 million of public funding
Key Points
- Cosine has announced Lumen Sovereign, described as Britain's first fully sovereign frontier AI model, backed by the UK Government's £500 million Sovereign AI programme.
- The coalition includes BAE Systems, Babcock International Group, BT, Lloyds Banking Group, LSEG, NatWest Group, PwC, Telefónica Tech UK&I, Thales UK and The Alan Turing Institute.
- The model will be trained entirely on UK soil using the Isambard-AI supercomputer and designed to run without dependence on foreign infrastructure.
- Cosine said roughly two-thirds of compute will target use cases including cybersecurity testing, KYC and AML investigations, and clinical trial coordination, with the rest focused on software engineering.
- The company is targeting delivery by late 2026.
AI firm Cosine has announced the formation of an industry coalition to co-design Lumen Sovereign, which it describes as Britain’s first fully sovereign frontier AI model.
The project is backed by the UK Government’s £500 million Sovereign AI programme. According to Cosine, the model will be trained entirely on UK soil using Isambard-AI, which the company describes as one of Europe’s most powerful supercomputers.
The coalition brings together a number of major UK institutions across defence, finance, telecoms and research. Cosine said the members are BAE Systems, Babcock International Group, BT, Lloyds Banking Group, LSEG, NatWest Group, PwC, Telefónica Tech UK&I, Thales UK and The Alan Turing Institute.
Cosine said the coalition members have signed a memorandum of understanding, taking what it called an active role in establishing a framework for defining the model’s use cases, security protocols and governance standards.
The company said Lumen Sovereign is being built to operate with no dependence on foreign infrastructure. Cosine said the model will be capable of deployment entirely within a customer’s own infrastructure, including fully air-gapped environments, with no external data transfer required.
According to Cosine, this addresses a choice that UK defence, financial and public sector institutions have faced until now: either falling behind in AI adoption or relying on models trained and operated overseas. The company said the latter can create what it described as unacceptable operational and sovereignty risks.
Cosine said a sovereign model offers advantages including domestic data control, alignment with UK and European regulatory requirements, and cost protection against vendor lock-in and price increases from foreign-built APIs.
On the technical approach, Cosine said it is using its allocation of UK sovereign compute to upcycle an existing open-weight model rather than pre-training a new one from scratch.
The company said this involves applying continual pre-training and expert expansion to a 256k-context sparse Mixture-of-Experts architecture, targeting a long-horizon agentic coding model within a 500,000 H100 GPU-hour compute budget.
Cosine said roughly one-third of the total compute time will be dedicated to software engineering problems, with the remaining two-thirds focused on use cases such as cybersecurity adversarial testing, KYC and AML alert investigations, and clinical trial coordination.
The company said it is targeting delivery of Lumen Sovereign by late 2026.
Writing in the announcement, Cosine co-founder and chief executive Alistair Pullen said the coalition was a step toward technological independence, describing the aim as ensuring Britain remains “an AI maker, not just an AI taker.”
