Cosine rallies UK leaders to develop a sovereign AI model.

      Banks, telecom companies, and arms manufacturers in Britain are united by a new concern: the AI systems they increasingly rely on are developed, owned, and managed in the United States. A startup that is not yet three years old is banking on their need to change that.

      Cosine, a pioneering AI lab based in the UK, has formed a partnership of prominent British organizations to co-create Lumen Sovereign, which it claims to be the country’s first sovereign frontier AI model.

      The list of participants features major names in the national economy, including BT, HSBC, Lloyds, NatWest, BAE Systems, Babcock, LSEG, PwC, Thales UK, Leonardo UK, and Telefónica Tech, all of which have signed a memorandum of understanding to help define the model’s applications, security standards, and governance.

      The announcement coincided with the launch of London Tech Week, during which Prime Minister Keir Starmer outlined a more proactive AI strategy and revealed approximately £400 million in new funding for specialized AI chips to enhance the nation's computing capacity. He emphasized that Britain’s future AI leaders should “start here, scale here and stay here.”

      The key advantage of this initiative is control. Lumen Sovereign will be trained solely on Isambard-AI, the Nvidia-supported supercomputer in Bristol, recognized as one of Europe’s most powerful, using resources allocated under the government’s £500 million Sovereign AI program, which selected Cosine in its initial group in April.

      According to Cosine, the model will have no reliance on foreign infrastructure at any point and can be implemented within the client’s own systems, including air-gapped environments that lack connections to external networks.

      This is the essence of their proposal.

      For many businesses, selecting an AI model is a straightforward purchasing decision; however, for defense contractors, banks managing anti-money-laundering processes, or operators of critical infrastructure, it swiftly becomes a matter of legal and security significance. Transmitting classified data, AML alerts, or clinical information to a server located in a US data center is frequently prohibited.

      “Businesses are becoming increasingly aware of the dangers of total reliance on foreign vendors,” stated Alistair Pullen, co-founder and CEO of Cosine, who pointed out that vendor dependency leads to “security risks, dependency issues, and rising costs.”

      Cosine may seem like an unconventional choice for this project, yet it is a credible one. Established in 2022 by Pullen, Yang Li, and Sam Stenner, this Y Combinator-supported lab has secured just $8 million from backers, including Lakestar, but its coding models have consistently outperformed those of OpenAI, Anthropic, Mistral, and DeepSeek in independent evaluations for two consecutive years. It accommodates over 38 programming languages, including COBOL, Fortran, and Ada, which continue to underpin Britain’s defense systems and financial infrastructure, and which are often poorly managed by most AI tools.

      Cosine asserts that Lumen Sovereign will be developed from the ground up using proprietary datasets covering over 30 regulated processes, rather than being refined from an open-source model, with a deployment goal set for late 2026.

      The coalition partners are candid about their involvement. "Cosine has provided us with a means to access a fully UK-native and highly customizable AI stack,” noted Peter Passaro, director of AI and data at Babcock, emphasizing “the very complex defense environments in which we operate.”

      Priority applications include cybersecurity, KYC and AML investigations, legal document analysis, and healthcare management—sectors where AI adoption in the UK has been hindered by these specific security issues.

      Cosine’s initiative represents the forefront of a broader British quest for AI autonomy, frequently running into a challenging reality. The government has invested funds into domestic computing and chip production and established sovereign infrastructure agreements with Nscale and Nvidia, even while key US projects like OpenAI’s Stargate UK have encountered setbacks.

      Nevertheless, "sovereign" AI in Britain typically still relies on American chips, including Isambard-AI.

      Currently, these are memoranda rather than binding contracts, and Lumen Sovereign has yet to be created. Training a frontier model from scratch to meet the assurance criteria required by the defense and finance sectors, and accomplishing this by the end of the year, is a daunting endeavor, especially given Britain’s past sovereign AI initiatives.

      What distinguishes this instance is the coalition: various institutions viewing sovereign AI not merely as a slogan but as a necessary procurement requirement.