Safe use and data

Gemini comes with Google Workspace business plans and answers from whatever the signed-in person can open. Where Workspace sharing runs wider than anyone meant, four checks to run and the controls Google gives an administrator.

Permissions decide what Copilot can reach, while sensitivity labels decide what it may do with it. What a label changes for Copilot, why labels do not replace permissions, a scheme small enough to use and what to label first.

Anyone with Copilot can build an agent, and Copilot Studio can build ones that act. Six decisions to make before your people build and share them, the controls Microsoft gives an administrator and where to start.

Copilot Chat comes with Microsoft 365 business plans at no extra cost and works on the file in front of you. The paid version adds your mail, Teams and files. What each does, what the free one protects and how to decide who needs a seat.

While an assistant answers and leaves the decision to you, an agent acts: it sends, updates, books and moves things. What that changes about permissions, errors and who can give it instructions, five questions before one goes live and where to start.

AI browsers read every page you open and can act inside your logged-in accounts. Why that is different, what Gartner advised in December 2025, four checks before one is installed and a sensible position for now.

AI note-takers record people who never chose them, and they spread through a calendar connection nobody approved. Five checks before one joins your calls, the built-in options and how to stop the ones you did not approve.

A new AI app arrives every few weeks, and some arrive with enormous attention. Six checks to run on any of them before work data goes in, a quick way to decide and what the DeepSeek launch showed about doing it in time.

Client information has gone into ChatGPT, probably on a personal account. What to do in the first hour, how to contain it, what you may owe and to whom, and how to stop it happening again, without blaming the person who told you.

A cloned voice or a deepfaked video call can ask for an urgent payment as convincingly as the real person. What happened at Arup, why spotting fakes is not the answer and the changes to payment approval that stop it.

AI states invented facts with the same confidence as real ones. Why it happens, where it shows up most, what it costs when it reaches a client or a court, what to check and the habits that make it rarer.

Data protection law did not change because AI arrived, and several of its questions are harder to answer for an AI tool. Eight checks before personal data goes in, what changed in 2026, and what belongs in the policy.

AI arrives in hiring through a feature in your recruitment system or a recruiter pasting applications into a chat tool. Why hiring is different, what regulators have found, seven checks before a tool screens candidates and where AI helps safely.

Which ChatGPT plan your people are on decides whether what they type can train the model, who controls the account and what leaves with them. What each tier changes, and how to move people onto one you control.

Most AI security writing is for the people who build models. For a business that uses AI tools, the risks sit in accounts, access, connections and payments. The six places AI changes your security and what to do first in each.

How to choose between Claude, ChatGPT and Microsoft Copilot for a business: the five questions that decide it, where each one tends to fit, and how to test them on your own work.