AI Daily Brief: 28 October 2026

28 October 2026

Today's news puts trust in the foreground: who is answering a consultation, whose data an AI service returns, and what an assistant is allowed to do. Alongside those warnings, Apple, Oracle and Opera are bringing AI closer to everyday business work.

Bots supplied 76% of responses to a UK vaping consultation

Research reported on Tuesday found that 90,835 of 118,756 responses to the 2023 tobacco and vapes consultation were automated. The Department of Health and Social Care identified and excluded those submissions, so the attempt did not stop the disposable-vapes ban. The findings concern a historical consultation, not a new attack this week.

Our take: Customer surveys, reviews and public consultations need checks on who is responding, not just how many responses arrive. Preserve participation for genuine people while monitoring duplicate content and unusual submission rates.

Cloudflare discloses cross-customer data exposure in one AI model

Cloudflare has published details of a now-fixed isolation failure affecting its GLM-5.3 Workers AI service between 27 August and 9 September. It identified 155 affected responses, all involving prompts longer than 681,280 tokens, and found no evidence of deliberate exploitation. A configuration mismatch and an SGLang inference-engine bug allowed old request data to appear in another customer's response.

Our take: Ask AI suppliers how they test tenant isolation at context limits and handle rejected requests. A model being accurate does not establish that its hosting infrastructure keeps customer data separate.

Apple brings an AI assistant into Pages, Keynote and Numbers

Apple is adding an assistant that can draft and restructure documents, refine presentations and organise spreadsheet data. The beta requires Apple Intelligence-compatible devices running iOS 27, iPadOS 27 or macOS 27 or later. On macOS 27, the apps also support Model Context Protocol connections for third-party AI agents.

Our take: Teams already using Apple productivity tools should test a real recurring task before adding another subscription. Check device eligibility, data handling and agent permissions before connecting business documents.

Oracle plans approved business workflows inside ChatGPT

OpenAI and Oracle have outlined Oracle AI Workspace, intended to let employees access approved Oracle data and complete business workflows through ChatGPT. The announcement also describes an Oracle AI SMB Bundle and App Composer. Access to OpenAI products through Oracle Marketplace is specified for US customers, so UK buyers should not assume identical purchasing availability.

Our take: The useful shift is from answering questions to working within business applications. Confirm what is actually available, which permissions carry through and where a human must approve a consequential action.

Opera previews local AI through its WebAI browser proposal

Opera has introduced WebAI, a proposed standard for running supported AI models on a user's own computer rather than relying on cloud inference. An opt-in early preview is available in Opera One, with other Opera browsers planned to follow. Supported examples include Google Gemma and Apple's on-device foundation model on Macs.

Our take: Local inference is worth testing for suitable private or repetitive work, but it is not a universal replacement for cloud models. Compare hardware requirements, model licences and output quality on your own documents.

Anthropic extends a 40% Claude discount across US government

Anthropic says its expanded General Services Administration agreement gives eligible US government organisations a 40% discount across Claude Enterprise, Claude for Government and the Claude API. It extends to state, local, tribal and territorial governments. The company says approximately 500 US government organisations use Claude and estimates $1bn in savings; that is a supplier forecast, not a measured outcome.

Our take: This is a US procurement agreement, not a UK discount announcement. For British buyers, the practical lesson is to negotiate around actual usage and implementation support rather than accept headline pricing as the whole cost.

Google DeepMind previews Genie 4 world simulation

Google DeepMind has announced Genie 4, an interactive world model with native audio, more realistic motion and improved physics simulation. It is initially a research preview for trusted testers, with wider access planned through Project Genie and early-access partners. Proposed uses include training environments for robots and agents as well as interactive experiences.

Our take: Treat this as emerging simulation technology, not a validated digital twin of your workplace. Training or safety decisions still require evidence that a generated environment reflects the real conditions that matter.

Frequently Asked Questions

How often is the AI Daily Brief published?

Every morning at 7:30am UK time, covering the previous 24 hours of AI news from over 30 sources.

How are stories selected?

UK-relevant stories are prioritised first, then by business impact and practical implications for UK organisations adopting AI.

Why should business leaders follow AI news?

AI is moving faster than any technology in history. Staying informed is essential for making smart decisions about AI investment, adoption, and governance.