Newsroom
What we are building, and why
Notes on work in progress. What each system is for, what problem it addresses and what phase it is in — without the implementation detail.
TissueForge: a million pounds of laboratory, or seventy-two hours of compute
Complex wound research is one of the slowest and most expensive corners of biomedical science. The arithmetic behind that sentence is worse than most people assume.
Read →CasualtyLogic: a hospital rehearses a major incident twenty times a year
In a mass casualty event, the decisions made in the first hours determine most of the preventable deaths. The tools for rehearsing those decisions run at human speed, one scenario at a time.
Read →PhishNet: sharing threat intelligence with people you have no reason to trust
Two organisations both being attacked by the same campaign have every reason to warn each other, and no shared infrastructure to do it through. That gap is where most of the damage happens.
Read →TrustVerify: detecting fakes is the wrong goal
Deepfake fraud took around €860 million in Europe last year, and attacks have risen more than twenty-fold since 2022. Chasing the fakes is a losing race. Certifying what is real is a different problem.
Read →HiveWire: we are not claiming an AI can make money. We are trying to find out
This is a research programme, not a product, and its own documentation forbids us from selling it until one specific thing is demonstrated. Here is what that thing is.
Read →InboxZeroAI: the one thing you cannot paste into someone else’s cloud
Your inbox is the most sensitive archive you own and the one most people hand over without thinking. InboxZeroAI is an attempt to sort it out without it ever leaving your machine.
Read →OctoSync: one agent is a bottleneck, many agents are a security problem
Running work in parallel is the easy part. Doing it without handing a language model the keys to the host machine is the part most systems skip.
Read →MirrorBreak: what happens when your AI stops disagreeing with you
A model that tells you what you want to hear is not being helpful. It is being useless in a way that feels like help — and the failure gets worse the longer the conversation goes on.
Read →ColdStart: the hardest moment with an AI is the first one
Most people leave the majority of a model’s capability unused. Not because the model cannot do it, but because they do not know how to ask. That gap is what ColdStart is being built to close.
Read →V.I.C.T.O.R.: a system that acts, not one that advises
Most assistants answer the question and leave the work with you. V.I.C.T.O.R. is being built to close that gap — and to keep working when the connection does not.
Read →Verifiable AI proof: the requirement that arrived before most companies were ready
European rules on artificial intelligence are in force. The organisations that lead will not be the ones using the most AI, but the ones able to prove they used it responsibly.
Read →