Sunday's reporting brings AI back to everyday business constraints: what each completed job costs, which connected systems need protecting, and whether new tools genuinely reduce people's workload. This shorter edition separates fresh reporting from recycled announcements and treats supplier promises as claims, not proven savings.
Fireworks argues AI buyers should measure completed jobs, not token prices
In a new Register interview, Fireworks executive Rob Ferguson argues that routing work between models can reduce costs compared with relying on one model. He says the company can halve bills for customers using a mixture of open and closed models, while warning that repeated processing and poor caching distort headline token prices.
These are a supplier's commercial claims, not an independent benchmark. For UK buyers, the useful comparison is cost per correctly completed task, including retries, human checking and data-handling requirements.
Our take: Take a representative set of real tasks and compare quality, latency and total cost before adding a routing service. A cheaper call is not a saving if staff spend longer fixing the result. Equally, dedicated hardware only makes commercial sense when utilisation and control requirements justify it.
London wholesaler's account shows the physical cost of insecure connected equipment
Since our previous reporting on the NCSC's AI-enabled attack warning, the Guardian has published a named business owner's account of recent refrigeration attacks. London wholesaler Alex Pearce describes disconnecting internet access after noticing abnormal defrost activity and temperature readings.
The report adds an important qualification: NCSC officials say attackers did not need AI to uncover the refrigeration weaknesses. Older equipment and known access weaknesses remain immediate business risks, regardless of how sophisticated future AI attacks become.
Our take: Inventory connected equipment as well as laptops. Check who maintains each controller, whether it still receives security updates and how staff can isolate it safely. Do not label every incident an AI attack: basic access controls and a rehearsed manual fallback still deserve attention first.
AI care assistants expose a familiar adoption problem: somebody still has to maintain the data
A new WIRED report examines tools designed to reduce family caregivers' administrative burden. Examples include Caring Village's Julia and Citizen Health's Ari, a beta assistant for people with rare diseases that can help organise records and prepare appointment questions.
The reporting also highlights a practical limitation: caregivers may still have to enter and maintain the information themselves. These are US examples, not evidence of NHS adoption or of availability and regulatory suitability in the UK.
Our take: For UK care providers and other service businesses, test the whole workflow rather than the chatbot alone. Establish who keeps records accurate, who checks generated summaries and whether the tool removes work or simply creates another system to update. Keep clinical decisions with appropriately qualified professionals.
US representative-elect sets out plans to seek a data centre moratorium
In a new WIRED interview, Michigan representative-elect Will Lawrence says he will fight for a federal data centre moratorium alongside a pause in advanced AI development. He links the proposal to safety, environmental protection and community concerns.
This is a politician's intended agenda, not an enacted restriction. For UK organisations buying cloud or AI services, it is another reason to examine where promised future capacity will actually be built and what delivery assumptions sit behind it.
Our take: Do not treat a supplier's announced campus as available capacity. Ask for service commitments, alternative locations and continuity arrangements. Planning consent, power availability and political opposition can all affect delivery, but this interview alone does not establish that any particular contract will be disrupted.
Nvidia reporting puts executive security and chip-customer checks under scrutiny
TechCrunch reports allegations from a US intelligence-based memo that a technology executive was recruited to target Nvidia chief Jensen Huang. Nvidia told the New York Times that his interactions with the executive were not unusual and that it does not share classified information with customers.
The report also points to a separate new investigation questioning the physical presence of an authorised Chinese chip buyer. The intelligence claims are allegations, not established findings of wrongdoing by Nvidia.
Our take: For UK businesses sourcing AI infrastructure, the practical issue is supplier and reseller due diligence. Verify the contracting entity, support arrangements and applicable export restrictions rather than relying solely on a familiar chip brand. This reporting is not evidence that ordinary Nvidia customers face an immediate service interruption.