TORUS
Hello Mate,
For UK founders, operators and professionals who want a fast, useful view of what changed in AI in the last 7 days.
Let’s dive in.
Today’s AI Signal
Britain opens a £3.1m route for frontier AI feasibility work
Innovate UK has opened a grant competition for UK-registered micro, small and medium-sized businesses. The pot is worth up to £3.1 million for feasibility studies in frontier artificial intelligence and machine learning technologies with a clear route to defensible scale-up. The opportunity was published on 5 October, opened on 1 October at 9:30am UK time and closes on 11 November 2026 at 11:00am UK time.
This is a single-applicant competition. The lead organisation must be a UK-registered SME, and the application needs to show a feasibility study rather than a general AI ambition. The useful distinction is between a model that looks promising and evidence that a business can use to make a larger technical or commercial decision.
Editor’s View: the strongest applications will define that decision before they define the system. A grant can pay for evidence. It cannot supply demand. Write around the buyer, the workflow and the result that would justify a larger build, not around the size of the model.
2. The MHRA is opening its AI Airlock to medical-device teams
The Medicines and Healthcare products Regulatory Agency has opened applications for phase 3 of its AI Airlock regulatory sandbox. Applications can be submitted year-round. The current review wave asks for submissions before 31 October 2026, with a later shortlisting period planned for spring 2027.
The focus is post-market surveillance and oversight across the lifecycle of AI-enabled medical devices. The programme describes simulated workshops, virtual testing with real or synthetic data, and real-world observation outside patient pathways. There is no application or participation fee. Selected teams are expected to fund their own studies and data access, and phase 3 is funded until April 2029.
3. Anthropic puts $100m behind the people who deploy AI
Anthropic says it is committing $100 million to train 10,000 Frontier Deployed Engineers by the end of 2027. The first cohorts include staff from Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey, Morgan Stanley and Novo Nordisk. Cohorts are running in San Francisco, New York and London.
The Frontier Deployed Engineer Residency starts with a simulated enterprise deployment, including use-case selection, security review and handover. It then moves into a 12-week residency built around a real Claude project inside the participant's organisation. The participant profile is hands-on software engineers with strong fundamentals and a record of building with LLMs; prior experience building AI agents is not required. The in-person stage includes a graded practical on a new scenario, followed by another assessment after 12 weeks. Participation is by nomination, and Anthropic expects the first graduates in early 2027. Organisations can ask their Anthropic account team or Partner Account Manager whether they are eligible.
Global AI headlines
TODAY’S LESSON
Turn an AI idea into a testable workflow
Before choosing a model or buying a tool, force the idea through one small operational test. The prompt below is designed for a founder or operator who has a messy process, a few examples and limited time.
You are a workflow analyst for a UK small business. I will give you a business process, its current steps, three representative examples, the people involved, the systems they use, and the risks we cannot accept. First, rewrite the process as a numbered sequence with one owner per step. Then identify the single bottleneck that creates the most delay or rework. Propose one AI-assisted change for that bottleneck only. State what the AI would receive, what it would produce, where a human must approve or correct it, and what data must not leave our approved systems. Define a seven-day test using no new paid software. Include a baseline measure, a target, a sample size, a failure log, and a stop condition. List two reasons the test could mislead us. Finish with a short decision rule: continue, change the workflow, or stop. Your output must include the numbered process, bottleneck, proposed change, controls, test plan, risks, and decision rule.Why it works: it turns a broad AI ambition into one measurable change with a human checkpoint. The stop condition protects the business from mistaking a polished demo for a useful process.
UK AI tools and offers
Offer: Frontier AI SME Champions, phase 1
UKRI lists this Innovate UK grant as open to UK-registered micro, small and medium-sized businesses, with up to £3.1 million available across the competition. It is a single-applicant route. Applications close on 11 November 2026 at 11:00am UK time.
Tool: Cohere North 2
Cohere describes North 2 as an enterprise AI platform bringing together security, intelligence, cost governance and full-stack control. The 5 October announcement positions it for deployment wherever an organisation's data lives. Torus verdict: ask where the platform runs, what governance controls are included and which evaluation data remains under your control before treating a platform claim as an operating advantage.
Tool: NVIDIA DGX Spark 64GB
NVIDIA says the 64GB DGX Spark configuration will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI on Friday 23 October 2026, starting at $4,999. It supports models up to 100 billion parameters on device, while two units can pool 128GB of memory through NVIDIA Sync Cluster Assistant. Treat the performance figures as vendor-reported and check the support, power and security model for your workload.
Tool: Claude Sonnet 5.5
Anthropic's 28 September release prices Claude Sonnet 5.5 at $2 per million input tokens, $10 per million output tokens and $0.20 per million tokens for cache reads. Anthropic reports that it runs more than 30% faster than Sonnet 5 and costs up to 30% less per task in its testing. Compare it with your own failure handling, review time and cost of being wrong.
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.
