TORUS
Hello Mate,
For UK founders, operators and professionals who want a fast, useful view of what changed in AI in the last 24 hours.
Let’s dive in.
Today’s AI Signal
Britain is moving from child safety consultation to product rules
The Department for Science, Innovation and Technology has published its response to the national consultation on children's online lives. More than 116,000 parents, children, teachers, clinicians, charities, campaigners and bereaved families responded.
The government says social media companies will not be able to provide their services to under-16s. Default protections for 16 and 17-year-olds would include restricted social-media access between midnight and 06:00, muted overnight notifications, and autoplay and personalised feeds switched off. The response also says the UK will ban sexualised AI chatbots for under-18s.
The government intends to put the first regulations before Parliament by the end of 2026, with restrictions expected to take effect in spring 2027. These are plans, not yet a complete operating rulebook. My view: AI suppliers serving UK consumers should map their age-assurance and child-access controls now, but should not pretend the implementation detail is settled.
UKRI opens £10 million programme for explainable AI research
UK Research and Innovation and EPSRC have opened a programme that will invest up to £10 million over two years in speculative, high-risk fundamental research on explainable AI. Applications officially opened on 11 August 2026.
Individual projects can have a full economic cost of up to £602,500. UKRI will fund 80%, up to £482,000. Applicants must be based at an eligible UK research organisation, and the call particularly welcomes early- and mid-career researchers. Projects must begin on 1 February 2027 and can run for up to 24 months.
Applications close at 16:00 UK time on 20 October 2026. The call is designed for research that could create a step change in explainability. It is not a general startup grant or a route for incremental product development.
AI evaluation should measure teams, not just autonomous models
A position paper accepted to the ICML 2026 Position Paper Track argues that the dominant approach to AI evaluation points development in the wrong direction. Jan Kulveit, Gavin Leech, Tomáš Gavenčiak and Raymond Douglas say benchmarks focused on superhuman autonomous performance implicitly treat replacing people as the objective.
The authors propose evaluating human-AI teams instead. That changes the question from whether a model beats a person working alone to whether the combined system produces better decisions, catches errors and handles work that neither side should own independently.
The paper is an argument, not evidence that one collaborative design already wins. Its practical consequence is clear: organisations should test the complete workflow, including human review and handoffs, rather than buying a model because it tops an autonomous benchmark.
Global AI headlines
TODAY’S LESSON
Turn the AI idea into a bounded experiment
Access to a model, grant or supercomputer is not evidence that a workflow creates value. Define the business decision first, then test the smallest dataset, baseline and failure condition that would justify another week of work. That discipline matters most when new frontier models arrive quickly.
Act as an operating partner for a UK small business.
We are considering this AI workflow:
[DESCRIBE THE WORKFLOW IN 3 TO 5 SENTENCES]
Turn it into a two-week experiment. Do not recommend tools yet.
Return:
1. The specific business decision or task being improved.
2. The current manual or non-AI baseline.
3. The smallest realistic test dataset, including what must be removed or anonymised.
4. One measurable success metric and a target threshold.
5. Three failure modes covering privacy, quality and operations.
6. A day-by-day plan for ten working days.
7. A stop, revise or proceed rule based on evidence.
8. The minimum human review before output reaches a customer, colleague or production system.
State every assumption. Flag missing information that could materially change the recommendation. Keep the plan practical for a team of fewer than 20 people and a limited budget.
Output specification: a one-page experiment brief with headings, a compact table, a ten-day plan and a final go or no-go checklist.Why it works: it separates the business question from the model question. The result can also support a sensible application for compute or funding because it defines the workload, evidence and stopping rule.
UK AI tools and offers
Tool: AIRR Rapid Access
AI Research Resource Rapid Access route is open to UK-registered businesses and is aimed particularly at startups and small and medium-sized enterprises. It provides up to 20,000 GPU hours for use within three months of the project start. Applications are reviewed monthly, with no closing date, and UKRI aims to provide access within two months of submission. This is compute, not cash funding.
Tool: AIRR Gateway
The AIRR Gateway route is open to UK-registered businesses of any size, charities, not-for-profits and eligible UK research organisations. It offers up to 10,000 GPU hours for use within three months. The route has no closing date and runs a monthly light-touch review. It suits first-time users, benchmarking and smaller proofs of concept.
Tool: EuroHPC AI Factories access
UKRI's EuroHPC access route lets UK startups, small and medium-sized enterprises, industry, public-sector teams and researchers apply for supercomputer time for AI and data-intensive work. UK applicants are eligible because the UK participates in EuroHPC. The call is open with no closing date. No cash funding is provided.
Offer: UKRI explainable AI research funding
UKRI opened a £12.05 million funding opportunity on 11 August for fundamental research into explainable AI. Projects can have a full economic cost of up to £602,500, with UKRI funding 80%, up to £482,000. Applications close on 20 October 2026.
This one is specifically for eligible UK research organisations, so it is not a general SME grant. Businesses working closely with university AI teams should still know it exists.
UK AI Job Opportunities
Citi - Lead AI Engineer, Senior Vice President, London, hybrid - posted 11 August 2026.
Citi - Senior AI Engineer, Vice President, London, hybrid - posted 11 August 2026.
Capgemini - AI Engineer, London - permanent Data & AI role posted 7 August 2026.
EDF - Data Scientist, AI/ML, UK hybrid - primarily home-based with London or Hove office travel. Applications close 23 August 2026.
Stay in the loop - TORUS tracks UK and European AI so you do not have to.
You are reading TORUS because you want to stay ahead of where AI is going, not just where it is today.
If you find this useful, forward it to one UK founder, operator or policy‑maker who should be thinking about AI infrastructure, tools and skills right now.
