We’re proud to welcome two new members to Cosine’s Lumen Sovereign coalition: QinetiQ and Xcelo.
By combining expertise in defence technology, we’re strengthening our mission to build a secure, sovereign AI capability for the UK.
Work on Britain’s first sovereign frontier AI, Lumen Sovereign, has begun. We’re opening a public request for input to shape the design, safety, and future of our national capability.
Backed by the @UKSovereignAI programme, Lumen Sovereign is being built on UK soil using the Isambard-AI supercomputer. It is designed to run in air-gapped, high-security environments, ensuring our most vital institutions – from defence and finance to public services – can innovate without compromise.
But we aren't building this in a vacuum. We’re working with a landmark coalition of industry leaders to define the governance and safety standards that will set the bar for the industry.
Now, we want to hear from you. We’re seeking insights on:
→ Which real-world use cases should we prioritise?
→ How can we better serve the workflows of UK institutions?
→ What safety and governance standards are non-negotiable?
Varied perspectives are vital. Help us build Britain's sovereign AI future.
Share your insights here: https://t.co/E9xTXTRsqz
The UK is punching above its weight in AI, third globally behind the US and China, with some genuinely great companies doing the work. Great to hear @CosineAI get a mention on BBC Radio 4 today. 📡
Start Rant
There is a peculiar class of people that emerges whenever our country attempts to build something strategically important. They rarely create the technology, rarely deploy it, rarely employ the engineers, and almost never accept responsibility for the consequences of inaction.
Their role is different. They appoint themselves custodians of progress, convinced that nothing of consequence should exist until it has first passed through an ever-expanding architecture of approval, certification, commentary and procedural purity.
The remarkable thing is not that these people exist. Every generation produces them.
The remarkable thing is how often Britain mistakes them for builders. We increasingly celebrate those who can explain why something should not be attempted over those prepared to shoulder the risk of attempting it.
We have built a culture in which criticism is treated as a substitute for creation, where the ability to identify imperfections is confused with the ability to produce excellence.
The rest of the world is not participating in this performance. America is building frontier AI because it believes technological leadership underpins economic and military power. China is building because it has reached the same conclusion. Other nations have looked at the world as it is, accepted that strategic capability matters, and got on with the difficult work of creating it.
Britain still contains too many people who seem to believe our comparative advantage lies in becoming the world’s referee rather than one of its competitors. They speak endlessly about standards, processes and frameworks, as though history has ever been shaped by the people who perfected governance while someone else perfected the technology itself.
The future will not be decided by those who positioned themselves closest to the microphone. It will be decided by those who built the capabilities that everyone else ultimately depended upon.
Every serious nation already understands this. Cosine intends to do whatever it takes to keep Britain relevant, resilience and grounded in reality. DM me if you want to be part of this journey. We’re hiring.
There's no better way to spend a minute and a half than listening to me talking about air-gapped AI deployments.
@CosineAI understands your codebase, writes pull requests, and ensures no sensitive data leaves your boundary.
Choose your model, and we’ll do the rest.
I recently discussed why sovereign AI is no longer optional on @BBCNews AI Decoded.
The US government’s decision to restrict access to GPT-5.6, following similar limits on Anthropic’s Fable and Mythos models, should be a wake-up call for UK enterprise.
What was a hypothetical risk is now a hard reality. When these systems are in active production, they can be switched off without a moment’s notice, leaving UK businesses with zero control.
We need resilience. @CosineAI is providing air-gapped models to guarantee continuity. We're also building Lumen Sovereign, the UK’s frontier AI model, with our public call for input going live this week. Keep an eye out to make sure your voice is heard.
https://t.co/OCpNCccoM4
Kicking off our time in Paris for #VivaTech with the @CosineAI team at the British Ambassador’s Residence. It's refreshing to see the UK & Europe talking seriously about AI – especially sovereign AI, as it’s more important than ever. Excited for the conversations ahead!
Infrastructure for the UK should be built with the UK.
To bring transparency to how we build a frontier AI model, we’re opening a public call for input on Lumen Sovereign in two weeks.
We want you to help shape the design, safety, and future of this capability. Stay tuned.
Over the weekend, a frontier model available to millions was switched off. This decision was made without input from any UK business, government, or developer.
The recent news about access restrictions on Anthropic’s Fable and Mythos models is a strategic wake-up call for every enterprise and government organisation.
If the intelligence your company or your country runs on is controlled elsewhere, you are borrowing it. When that access is throttled, gated, or revoked overnight, it becomes a strategic vulnerability.
As our Co-Founder @yangli_ said earlier last week, “AI capability is now strategic national infrastructure.”
We are building Lumen Sovereign so Britain has a frontier AI model it actually owns: trained here, run here, and designed to operate beyond the reach of external control.
This is exactly why we @CosineAI are using the compute provided to us by @UKSovereignAI to build a large frontier model trained in Britain.
This export control is the first instance of it’s kind but, I think it’s going to become increasingly common as the models improve.
Regulated industries are often the most restricted by standard models. That’s why Cosine is building Lumen Sovereign: a model designed for regulated UK industries to unlock secure, compliant intelligence.
Find out our approach to what sovereign AI can unlock.
Great to see @CosineAI featured in @NVIDIA’s blog on UK sovereign AI advancements! As a member of the #NVIDIAInception program, it’s exciting to be part of the growing UK AI ecosystem and drive innovation.
https://t.co/cu7V3nCbFl
"Swarm mode was great!"
@andrewkuanop won the Cosine prize at yesterday's @AntlerGlobal hackathon after building marketing tool, Cast, in just a few hours.
Here's what the team built & how @CosineAI helped✨
We are incredibly proud to have @BAESystemsplc supporting our landmark industry coalition. Together, we are co-designing Lumen Sovereign to ensure the UK has an unshakeable, self-reliant AI stack built for our most critical defence environments.
Why does the 🇬🇧 need its own 'frontier AI' model?
For defence, it means greater resilience and stronger data security. That's why we support @CosineAI's work on a complete UK AI stack, built on UK soil. Even if satellite links or undersea cables were cut, sovereign AI would keep the UK operational.
Read further insights from our technology director, Robert Merryweather💡 https://t.co/W6PRraRbKK
Today, Cosine hosted a breakfast panel in Westminster on Britain’s AI future, discussing how we can move from "takers to makers" through procurement reform and strategic investment.
Huge thanks to our panelists and Lord Ed Vaizey for the insightful discussion.
Big day for us Cosine, very excited to announce this publically and, again, thanks for Sovereign AI for helping make this possible.
Lumen Sovereign will be built with input from some of the largest companies in Britain including HSBC, BAE Systems, Babcock International Group, BT Group, NatWest, LSEG, PwC, Barclays, Telefónica Tech, Thales, Lloyds Banking Group with more to be announced soon.
Cosine is an AI coding agent for real software engineering.
It scales from quick fixes to complex migrations, working in your codebase via CLI or Cloud.
Available on macOS, Linux, and Windows.
Gemini 3.5 Flash is now available in Cosine.
With impressive response times, our engineers have found that the model is best suited for high-frequency code edits, rapid-fire debugging, and swift repository-wide exploration.