Britain's AI growth forecasts are getting a reality check: adoption could create substantial value, but the assumptions matter as much as the headline number. Today's developments also put evidence, security and the true cost of automation firmly on the business agenda.
Related reading: Why AI Evaluations Are Replacing Demos in Enterprise Buying Decisions
BCC says wider AI adoption could lift UK GDP by more than 10%
New British Chambers of Commerce modelling suggests UK GDP could be more than 10% higher by the end of 2031 under a high-adoption scenario. Its moderate scenario produces an uplift of around 2%, illustrating how heavily the outcome depends on implementation.
The higher case assumes AI boosts productivity by 57% across 12% of work tasks by 2030. The BCC is calling for a business adoption strategy built around skills, investment and practical support, rather than infrastructure alone.
Our take: This is a conditional forecast, not growth already banked. For a business owner, the useful exercise is smaller: choose one recurring task, measure its current cost and test whether AI improves the finished result after checking and rework.
WIRED traces the UK's £400 billion AI claim to Google-funded research
WIRED reports that the £400 billion economic benefit figure repeatedly cited by UK officials came from a 2023 Public First report commissioned by Google. AI minister Kanishka Narayan has set it alongside a government assessment of £55 billion to £140 billion by 2030, with a central estimate of £120 billion.
The government describes its estimate as highly uncertain. The consultancy says slower adoption helps explain the difference, and maintains that its research judgements are independent.
Our take: Do not treat forecasts with different assumptions and time horizons as interchangeable. Apply the same scrutiny to supplier ROI claims: ask who funded the study, what adoption it assumes and whether the calculation includes training, integration and human review.
Forrester predicts AI will influence 75% of B2B buying decisions by 2027
Forrester expects AI agents to influence three-quarters of business purchasing decisions by 2027, according to reporting by IT Brief. That includes discovering suppliers, gathering information and analysing contract renewals, not simply allowing software to buy autonomously.
The forecast also anticipates buyers doubling human interactions to validate AI recommendations. For suppliers, consistent product information and verifiable evidence become more important, not less.
Our take: Treat this as a forecast, not a measurement of today's customers. Audit the information a buyer can actually verify about your business: prices, scope, limitations, case evidence and named responsibility. AI-readable content still needs a person who can stand behind it.
OpenAI reports 948 Linux privilege-escalation exploits from an automated pipeline
OpenAI says a research pipeline using an upcoming cyber model dynamically reproduced 1,724 Linux vulnerabilities and generated 948 working local privilege-escalation exploits during a mid-October run. It reports an API cost of $1,213,430, roughly $1,279 per exploit.
The company says human researchers reviewed the outputs and that disclosure and candidate patches are being coordinated with Linux maintainers and Patch the Planet partners. These findings are not a claim that every affected production system has already been fixed.
Our take: The business implication is a shorter gap between finding a weakness and being able to exploit it. Ask your IT provider about supported operating systems, patching times, least-privilege access and recovery testing. A model-generated patch is not a substitute for a reviewed vendor update.
Cognition launches SWE-2 Flash with a lower-cost coding pitch
Cognition has released SWE-2 Flash in Devin Desktop and CLI, claiming three times the speed and one-fifth the price of SWE-2. It is based on Thinking Machines Lab's Inkling-Small, with 276 billion total parameters and 12 billion active parameters.
Its own benchmarks show trade-offs: Flash scores 24.0 against SWE-2's 23.8 on FrontierCode 2.0, but trails on Terminal-Bench 4.0 and DeepSWE 1.1. Cognition is developing controlled access to weights for partners and researchers rather than releasing them publicly.
Our take: Cheaper models make routine development work more affordable, but cost per accepted change matters more than cost per token. Compare models on representative tasks and include review time, failed attempts and maintenance before changing your development budget.
Anthropic reportedly licenses AP election data for Claude
ADWEEK reports that Anthropic agreed a content and data licensing arrangement with the Associated Press shortly before the US midterm elections. Claude users were shown up-to-date election information attributed to AP.
The report describes it as Anthropic's first licensing deal with a digital news publisher. Financial terms and duration were not disclosed, and the article says neither party provided an on-record confirmation of the arrangement.
Our take: Reliable answers require access to reliable current information, not just a capable model. When deploying an assistant, establish which sources it can use, how fresh they are and whether you have permission to use them. Attribution and a clear update process should be part of the specification.
Google credit backing supports TeraWulf's Anthropic campus financing
TeraWulf has announced Google credit support for certain Anthropic lease obligations at its Kentucky Justified Data Campus, according to reporting that links the company announcement and regulatory filing. The arrangements support anticipated financing tied to around $19 billion in contracted lease commitments.
The campus covers approximately 401 megawatts of critical IT load. Phased service is expected from late 2027 into the second quarter of 2028; debt financing and construction pricing are not yet fully settled.
Our take: The distinction between contracted capacity, financed construction and operating infrastructure matters. UK buyers should focus on the capacity and service guarantees available now, while checking concentration risk when their model provider, chip supplier and infrastructure backer are closely connected.