ODI tested a National Data Library prototype with 100,000+ datasets. Poor labelling, stale records, broken access throughout. AI defaults to news articles when gov data is unusable. https://resultsense.com/r/2026-03-31-government-datasets-poorly-labelled-for-ai #OpenData

OpenAI open-sourced its AI behavioural spec. Concrete decision rules, not vague principles. Worth reading if you're building governance policies. https://resultsense.com/i/2026-03-31-openai-model-spec-ai-safety-governance-framework #AIGovernance

Nscale's planned AI data centre in Essex faces local backlash after energy demands grew 50% beyond the original proposal. Project delayed to 2027. https://resultsense.com/r/2026-03-30-nscale-data-centre-faces-energy-backlash-essex #AI #Energy

Britain's founded more AI companies per capita than any other European country since 2020. London's now drawing talent from the US and across the continent. https://resultsense.com/r/2026-03-30-london-ai-sector-talent-synthesia #UKTech #AI

Manchester's largest NHS trust is deploying 6,500 Microsoft Copilot licences and building an 'Agent Factory' for AI automation. HR pilots already cut admin time in half. https://resultsense.com/r/2026-03-30-manchester-nhs-trust-expands-microsoft-ai-tools #NHS #AI

Nearly 700 real-world cases of AI chatbots scheming and ignoring instructions. UK-funded study found reports surged five-fold since October 2025. https://resultsense.com/r/2026-03-30-ai-chatbots-ignoring-human-instructions-study #AISafety

Oxford and UK AISI researchers prove AI models can escape Docker containers through common misconfigs. Cost per attempt: ~$1. https://resultsense.com/i/2026-03-30-sandbox-escape-bench-llm-container-security-benchmark #AISafety

New Deep Dive episode: Institutions Adopt, Controls Lag Behind 🎙️ NHS trusts ordering thousands of AI licences, HSBC's first Chief AI Officer, and 700 cases of chatbots going off-script. Big week. https://resultsense.com/podcast/s01e20 #Podcast #AI

Theia Insights raised £6.3m to replace static industry labels with real-time AI classification. Their data: firms with meaningful AI exposure grew from 63 to 2,900+ since 2013. https://resultsense.com/r/2026-03-27-theia-insights-series-a #UKTech #DeepTech

NYC hospitals are dropping Palantir when the contract expires in October. Staff had access to patient health notes beyond research. The NHS's £330m Palantir deal faces the same questions. https://resultsense.com/r/2026-03-27-new-york-hospitals-palantir-ai #NHS #DataPrivacy

A US judge blocked the Pentagon from punishing Anthropic for refusing unrestricted military AI use. First legal precedent for AI companies drawing ethical red lines on their products. https://resultsense.com/r/2026-03-27-anthropic-pentagon-court-victory #AISafety

Barnsley Hospital starts AI pilots in April to tackle waiting lists and no-shows. First output from the UK's Tech Town programme, with an £800k upskilling fund for local businesses in May. https://resultsense.com/r/2026-03-27-bbc-ai-hospital-waiting-times #NHSAI #TechTown

AISI studied 177,000 AI agent tools. 65% now take real-world action; just 27% did 16 months ago. Most governance frameworks haven't caught up. https://resultsense.com/i/2026-03-27-ai-agent-usage-evidence-177000-tools #AISafety #AIAgents

AI is now a national security sector in UK government procurement, alongside steel and shipbuilding. British suppliers get priority for public contracts. https://resultsense.com/r/2026-03-26-uk-government-prioritises-ai-procurement-national-security #UKAIPolicy #GovTech

M&S is deploying 11,000 Copilot licences to all store managers for sales insights, handovers, and trading summaries. One of the larger confirmed AI rollouts in UK retail. https://resultsense.com/r/2026-03-26-marks-spencer-rolls-out-11000-copilot-licences #RetailAI #UKBusiness

The UK's first tech town is putting AI to work. Barnsley Hospital gets an AI pilot to cut waiting lists, and there's an £800k fund to train local SMEs in AI skills. https://resultsense.com/r/2026-03-26-barnsley-tech-town-nhs-ai-upskilling #NHSAI #UKTech

Chinese AI labs charge $2-3 per million tokens. Western providers? $15. Agentic AI burns 20 million tokens per task, so that sixfold gap is now a strategic problem for UK firms. https://resultsense.com/i/2026-03-26-china-ai-token-economy-global-competition #AIStrategy #UKTech

Revolut made £1.7bn profit but flagged AI and crypto as reputational risks. Energy costs are the worry, right as it starts offering UK current accounts. https://resultsense.com/r/2026-03-25-revolut-ai-crypto-reputational-risk-record-profits #Fintech #AI

Feedzai's built a foundation model for financial crime, trained on £7.8 trillion in annual payments. Lloyds is an early partner. https://resultsense.com/r/2026-03-25-feedzai-riskfm-foundation-model-financial-crime #Fintech #AI

HSBC has appointed its first Chief AI Officer. NatWest, Barclays, and Lloyds are all restructuring around AI too. UK banking's organisational shift is real. https://resultsense.com/r/2026-03-25-hsbc-appoints-first-chief-ai-officer #UKBanking #AI

GOV.UK Chat pilot is done. 10,000 users, 26,000 questions, accuracy up from 76% to 90%. All 508 jailbreak attempts blocked. Wider rollout coming this year. https://resultsense.com/r/2026-03-25-gov-uk-ai-chat-completes-public-pilot #GovTech #AI

Hundreds of AI experiments across UK government. Zero that have cut costs. And 37% of citizens see it as a risk, not an opportunity. https://resultsense.com/i/2026-03-25-state-of-the-state-2026-uk-government-ai-digital-transformation #UKGov #AI

DVLA is deploying LLM-powered search across 2,500 knowledge articles and piloting AI voice for government call centres with GDS. 1.1m customer queries a month. https://resultsense.com/r/2026-03-24-dvla-ai-customer-service-transformation #AI #GovTech

Bank of Ireland is training every employee in AI by end of 2026. The numbers: €9.7m in fraud prevented, 40% fewer call transfers, 72% of customers say banks should invest in AI staff training. https://resultsense.com/r/2026-03-24-bank-of-ireland-ai-workforce-training #AI #Banking

The FCA is paying Palantir £30k a week to run AI on its data lake of case files, fraud reports, and consumer complaints. Legal experts say the data includes emails and recordings from innocent people. https://resultsense.com/r/2026-03-24-fca-palantir-ai-data-trial #AI #FCA

A Guardian writer let ChatGPT plan a London day. Brilliant picks; forgot food, rest, and basic human needs. That gap between AI recommendations and lived reality matters. https://resultsense.com/i/2026-03-24-ai-guide-london-consumer-experience-reality-gap #AI #UX

Google is replacing publisher headlines in search results with AI. One Verge headline had its critical angle stripped entirely. They call it an "experiment." Last time, that became permanent within a month. https://resultsense.com/r/2026-03-23-google-search-ai-rewriting-news-headlines

Palantir now has access to the FCA's case files, fraud reports and phone recordings. £30k+ per week for a three-month AI pilot. The government says it wants British suppliers. Interesting timing. https://resultsense.com/r/2026-03-23-palantir-wins-fca-contract-sensitive-financial-data

Starling Bank has launched the UK's first agentic AI for personal banking. It sets budgets, tracks spending and acts on your behalf. Not just a chatbot. 8 years in development, built on Google Gemini. https://resultsense.com/r/2026-03-23-starling-bank-launches-agentic-ai-money-manager

North Wales is using AI to triage cancer biopsies. Patients who waited 3 months for a diagnosis are now getting results in 1-2 weeks. Builds on earlier prostate and breast cancer AI pilots. https://resultsense.com/r/2026-03-23-welsh-ai-cancer-detection-trial-results

HM Land Registry's AI processed Newham's land charges in 4 weeks with 4 staff. Manual way: 3 months, 20 people. Every batch passed QA first time. 5 years to build. Patient engineering. https://resultsense.com/r/2026-03-23-hm-land-registry-ai-speeds-local-land-charges-migration

The AI adoption gap isn't between companies; it's inside them. Some teams use it daily, others haven't tried it once. Three experts have a useful rule: only use AI where you know what good looks like. https://resultsense.com/i/2026-03-23-guardian-expert-guide-using-ai-tools-effectively

New episode of The Deep Dive. UK spending billions on AI and quantum; safety commitments getting quieter. The pattern is clear. https://resultsense.com/podcast/s01e19

Two years ago, GPT-4 cost $33 per million tokens. Cheapest model now: 9 cents. Jensen Huang says tokens are AI's new currency, but nobody's proven the link between cheap tokens and business returns. https://resultsense.com/r/2026-03-20-nvidia-token-economics-ai-industry #AI #Nvidia

A landlord cited AI-generated law in Kirkcaldy Sheriff Court. The Act didn't exist. Neither did the case references. Sheriff warned of contempt. Believed to be a first in Scots law. https://resultsense.com/r/2026-03-20-scottish-judge-warns-ai-fake-law-contempt #AI #LegalTech

Criminals using AI for fraud earn 4.5x more, Interpol finds. Deepfake kits now cost dollars on the dark web. Agentic AI could soon run autonomous scam campaigns end to end. https://resultsense.com/r/2026-03-20-interpol-ai-fraud-four-times-more-profitable #AI #CyberSecurity

Meta's AI agent gave flawed guidance that exposed sensitive data to staff for two hours. Amazon's had similar issues. Agentic AI without proper access controls is a real risk. https://resultsense.com/r/2026-03-20-meta-ai-agent-triggers-sensitive-data-exposure #AI #DataSecurity

Anthropic asked 80,508 AI users what they want. Only 18.8% said productivity. Most want growth, time freedom, meaning. Big question for AI strategies that only measure efficiency. https://resultsense.com/i/2026-03-20-anthropic-81k-ai-aspirations-global-study #AI #UKBusiness

UC San Diego researchers say multi-agent AI has a memory problem. The fix? Computer architecture: caching, consistency, hierarchy. Old problems, new systems. #AI #AgentSystems https://resultsense.com/i/2026-03-19-multi-agent-memory-computer-architecture-perspective

PwC's US chief has told partners: embrace AI or leave. The firm's launching an automated platform replacing hourly billing with subscriptions. Big Four model is changing fast. #AI #BigFour https://resultsense.com/r/2026-03-19-pwc-ai-overhaul-partners-resist-technology

UK government's dropped its preferred AI copyright approach after backlash from artists including Elton John and Dua Lipa. Four options remain, no timeline. #AIPolicy #UKTech https://resultsense.com/r/2026-03-19-uk-government-backtracks-ai-copyright-creative-industries

Visa's launched AI payment trials with Barclays, HSBC UK, Nationwide, and Revolut. Testing how AI agents can securely buy on consumers' behalf. Mastercard ran similar trials with Santander weeks ago. https://resultsense.com/r/2026-03-18-visa-launches-agentic-payments-trials-with-uk-banks

EU Council proposes pushing AI Act high-risk deadlines back up to 16 months. New dates: December 2027 and August 2028. Standards needed for compliance still haven't been published. https://resultsense.com/r/2026-03-18-eu-council-proposes-delays-to-ai-act-high-risk-rules

Trustpilot's AI search click-throughs grew 1,490% year-on-year. It's the fifth most-cited domain on ChatGPT globally. Now pursuing e-commerce partnerships to monetise the position. https://resultsense.com/r/2026-03-18-trustpilot-targets-ai-driven-growth-with-ecommerce-partnerships

Skills England launched a Level 4 AI apprenticeship open to all sectors. First cohort's in, from tourism to addiction charities. Part of a push to upskill 10m workers by 2030. https://resultsense.com/r/2026-03-18-uk-launches-ai-apprenticeship-to-upskill-millions

Chancellor commits £2.5bn to AI and quantum. A £500m Sovereign AI Fund launches in April, plus £2bn for quantum computing including the UK's first commercial quantum procurement programme. https://resultsense.com/r/2026-03-18-chancellor-pledges-2-5bn-for-uk-ai-and-quantum

AISI tested frontier AI on a 32-step corporate network attack. Best run: 22 steps completed. Eighteen months ago, top score was 1.7. Security planning needs to catch up. https://resultsense.com/i/2026-03-18-frontier-ai-agents-multi-step-cyber-attacks-aisi-evaluation

GovAI/Brookings research: 6.1m clerical workers are most exposed to AI and least able to adapt. Women hold 86% of these roles. The UK has similar patterns. #AIjobs https://resultsense.com/r/2026-03-17-which-jobs-are-most-threatened-by-ai-new-research-maps-the-risks

US Treasury published an AI risk framework with 100+ financial institutions: 230 controls across four governance functions. UK firms, take note. #AIgovernance https://resultsense.com/r/2026-03-17-us-treasury-publishes-ai-risk-framework-for-financial-services

New Law Society guide on AI for solicitors: genAI hallucinates during strict data extraction. For contract review, purpose-built ML tools are more reliable. #legaltech https://resultsense.com/r/2026-03-17-how-solicitors-are-using-ai-for-contract-review