Introducing Sukarma
Three years ago we set out to build a company for the moment AI started writing software. That moment has arrived. Here is what we built while we were quiet, and why we are focusing on public services.
Read articleThe AI Playbook for the UK Government sets out ten principles for safe and effective AI. Here is how we turn each one into something concrete in our delivery.
The AI Playbook for the UK Government, published by the Government Digital Service in 2025, sets out ten principles for using AI safely, effectively and responsibly. It is written for civil servants, but we think every supplier to the public sector should read it as a specification.
Here is how we put the principles into practice.
Every proposal we write states plainly what the AI component will and will not do, how accurate we expect it to be, and how we will measure it. We do not sell certainty that the technology cannot deliver.
We screen every use case for data protection, equality and ethical risk before building, and produce a draft data protection impact assessment and equality considerations alongside the first prototype.
We design against the NCSC guidelines for secure AI system development and test against the OWASP Top 10 for LLM Applications, including prompt injection, sensitive information disclosure and excessive agency.
Our systems are built with explicit approval gates. Consequential decisions are made by named people, and the interface makes it easy to see what the AI did and why.
Models, prompts and evaluation sets are version-controlled. We monitor quality and drift in live services and can roll back a model change as easily as a code change.
Sometimes the right answer is a rules engine, a better form or a simple search box. Our discovery work considers non-AI options first, and says so when they win.
We favour open standards and open interfaces, and we draft Algorithmic Transparency Recording Standard (ATRS) records for the tools we deliver.
We price transparently, explain model running costs up front and avoid contractual lock-in, so commercial teams can make informed decisions.
We leave knowledge behind: documentation, runbooks and pairing with in-house teams, so the organisation can own and evolve what we build.
We fit into your governance, not the other way around: your security policies, your design authority and your assurance boards.
The Playbook is a good document. The test of a supplier is whether they can show you how they meet it.
Keep reading
Three years ago we set out to build a company for the moment AI started writing software. That moment has arrived. Here is what we built while we were quiet, and why we are focusing on public services.
Read articleWhy we are called Sukarma, why our symbol is a ripple, and the five promises behind them.
Read articleGoogle has expanded its SynthID Detector worldwide and the EU's transparency rules for AI-generated content are coming into effect. Public bodies should be clear about where they use AI.
Read article