AIVS Software

Nominated by Alexandre Maurice, AIVS Software

AIVS has built a sovereign AI platform that can run on an organisation's own premises or fully offline. Every answer is grounded in approved sources, with the supporting passages shown, and human review is recorded so that decisions can always be traced back to a person.

Its crisis-management product, CIMA, grades risk using fixed rules rather than the AI model's own confidence, and keeps operational knowledge separated by domain; a sister product, PRIMA, serves accountancy. AIVS works with contracted defence customers, and in a published trial a verified, fully cited answer took 10 minutes against about two hours by hand.

EVE.CALLS

Nominated by Angela Stathi, Office for Investment

EVE.CALLS builds voice agents that make and answer calls for banks, lenders and public services. Every call follows protocols approved by the client, hands hardship cases and disputes to a member of staff, and is re-checked afterwards by a separate automated auditor.

At a state-owned bank in Ukraine, a review of a 10% sample of calls found every conversation accurate and within policy, with 99% accuracy on recorded payment commitments. The platform is SOC 2 Type II audited, and its founder co-authored Ukraine's voluntary code of conduct on the responsible use of AI.

Great Wave AI

Nominated by Tom Cotterell, Great Wave AI

Great Wave AI's platform for building and running AI agents has three safeguards that cannot be switched off: every interaction is scored in real time for accuracy and relevance, staff can give feedback directly in the platform, and every question, answer and source is fully audited.

More than 1,000 of these governed agents run in live public services at the Ministry of Defence, the Insolvency Service, the Department for Business and Trade and Warwickshire Police. The platform runs inside each customer's own infrastructure and is ISO 27001 certified.

Responsible AI UK

Nominated by Gopal Ramchurn, Responsible AI UK

Responsible AI UK is a national programme hosted at the University of Southampton and funded by UKRI. It brings together 28 universities and more than 150 partners, and has funded 42 projects to make AI more accountable and trustworthy.

Its practical tools include an accountability toolkit for policing now being reviewed by five forces, a responsible AI checklist for policing, and an open-source workbench for auditing AI harms developed with 28 partners including Microsoft and the Information Commissioner's Office. Its evidence has informed Parliament, the Ministry of Justice and the International AI Safety Report.

Tech Against Terrorism

Counter-Terrorism AI Benchmark

Nominated by Cat Cadenhead, Tech Against Terrorism

Tech Against Terrorism's Counter-Terrorism AI Benchmark is the first systematic test of how AI models can be misused by terrorists and violent extremists. It evaluated 27 models, including open-weight versions with their safety guardrails removed, against almost 2,400 prompts drawn from real cases.

The methods, the classification of misuse and the full results are published openly, with findings shared with AI providers first. The benchmark launched at the UN's Counter-Terrorism Week and has been briefed to parliamentarians.