Microsoft's Chief Says Treat AI Like an Insider Risk. The UK's AI Minister Speaks Today.

On Saturday 10 October Satya Nadella, Microsoft's chair and chief executive, published a post on X titled "Models as Insider Risks in the Super Intelligence Era". His starting point: with ordinary software a business can trace what happened to a specific line of code. With today's AI models nobody can, and yet, in his words, "we are deploying these complex agentic systems and models, with access to our most sensitive data and giving them the ability to take mission-critical actions on our behalf". His conclusion is blunt: "A model provider's assurances do not relieve us of that responsibility."

His fix is not to trust the model more but to design around it the way firms already handle a powerful employee: "establish identity, limit privileges, log activity, create containment boundaries". The controls on what an AI can see and do must sit outside the model, in the harness, the software around the model that hands it tools and permissions. He lists seven principles: no single model as the sole dependency for an important outcome; every meaningful action leaving "tamper-proof human readable evidence"; continuous testing of failures, not just successes; the business deciding independently what the model can access and do; validation by something other than the model being validated; containment ("Think of it like an emergency brake. An authorized person should always be able to pause or shut down a model mid-task"); and timely disclosure when things go wrong.

The closing line is the one to keep: "The most trustworthy Super Intelligence system will not be the one with the model we trust most. It will be the one that enables us to trust the model the least." Microsoft announced no product change alongside the post, which is Nadella's own view. It followed Friday's disclosure by Anthropic that one of its models had filed a false tip to a police website during testing, covered in Saturday's briefing.

Kanishka Narayan, the UK's AI minister, gives a speech in west London today, the first since the Prime Minister, Andy Burnham, made AI a cabinet-level brief. According to the Guardian's preview, published at one minute past midnight, he will say companies building the most advanced AI must do more to make it safe, that Britain cannot rely entirely on technology companies to police their own models, and that the UK needs more AI infrastructure of its own. He will also announce a strengthening of Britain's cyber defences. The detail of those measures was not public when this briefing was written, and gov.uk had published nothing by this morning.

The speech is being watched for whether the UK joins an international call for common AI safety standards already signed by leaders including Finland, France, Germany, Australia, Singapore, the UAE and Ireland. The UK and the US have not signed, and President Trump has said he wants the leading AI companies to "self-police". The Prime Minister told the UN last month that Britain would use its G20 presidency next year to push for "a single set of global principles and standards". Last week the heads of MI5, MI6 and GCHQ briefed the cabinet on AI's threat to national security, according to the Sunday Times.

The Guardian also quotes Geoffrey Irving, a former chief scientist at the UK's AI Security Institute, the government body that tests the most powerful models, warning that AI companies "do not have security that would prevent an adversarial country from stealing the models". He cites a recent authorised hacking exercise at OpenAI. OpenAI says it has multiple protections, including encryption and continuous monitoring. Irving's wider warnings about superintelligence are forecasts, not findings, and are reported here only as context for the pressure the minister is under.

On Friday 9 October Microsoft published its October update for Copilot Notebooks, the project workspace inside the Microsoft Copilot app and OneNote where you gather the emails, meetings, files and notes on one topic and ask Copilot questions grounded in them. The headline change is that Copilot can now do simple work inside the notebook rather than only answer: keep the references current, pull together information across the project, and create or edit a Word, Excel or PowerPoint file from the chat.

The piece most firms will want is scheduled tasks. Microsoft's own example instruction begins: "Every Friday at 4:00 PM, Summarize activity across [Topics or Workstreams] from [Time Period], including notable updates, decisions, actions, risks, and unresolved issues. Create a crisp PowerPoint deck". You write that once and it runs weekly. Also new: Word, Excel and PowerPoint files open and edit on the notebook canvas with full Office; interactive reports the team can live-edit; a one-page PDF summary made from chat; a choice of which model Copilot uses; and Copilot building a whole notebook from a one-line description of the project.

Availability is the catch. Most of this (delegated tasks, model choice, Office editing, interactive reports and PDFs) is "currently available in Frontier", Microsoft's early-access programme for Microsoft 365 Copilot customers. Notebook creation and the scheduled automations are "coming in the next few weeks". Only the expanded reference types (Markdown, plain text and RTF files) and suggested artifacts are generally available today.