Office management edition
Fake invoices and deepfake requests are becoming impossible to spot, while Teams is starting to look up answers on its own and Gemini remembers where you put your passport.
The old spelling mistake is gone: here's how to check a request that comes across your desk
AI makes fake messages and fake voices so real that the safety net role now falls on you. That calls for a fixed checking routine, not gut feeling.
Cybersecurity experts warn that the usual signs of phishing no longer work. Spelling mistakes, clumsy sentences, strange addresses: AI-made messages no longer have them. They are often impossible to tell apart from real communication. Criminals go further than text. They use deepfake audio and video to imitate managers, to get payments approved, or to pressure team members. Security training must therefore go beyond the classic phishing example, with deepfake exercises and fixed verification agreements. This affects your work directly: an urgent invoice, a supplier with new bank details, a director calling from abroad, it lands with the office manager, often in a rush. Credit insurer Allianz Trade reports that invoice fraud already hit 44 percent of 350 organisations studied in early 2026, and has topped the list of external fraud for years. This week, agree one fixed verification rule with your manager for every request for money or data, and put it in writing. Why did the old rule work for so long? Fraudsters made language mistakes, so the mistake was in the language. An AI text model writes flawless Dutch and can even copy a company's tone. So recognition shifts from the shape of the message to the route it takes. The same goes for a deepfake: the voice sounds right, so the check must sit in the channel. Call back on a number you already had, not the number in the email. KVK gives the same advice for invoices, and it also works for an urgent phone call. Calling back on a request from management can feel suspicious, but it only works if it is policy, not your own personal choice. Then it is not distrust of the director, it is a rule that applies to everyone. Watch out for false confidence too: deepfake training makes you alert, but no one can reliably spot a good fake voice. The difference lies in the process, not in your ear.
Background for subscribersThis edition shows that the old rules of thumb no longer work. AI tools make fake invoices and fake voices impossible to tell apart from the real thing, so your fixed verification routine becomes the real protection. At the same time, tools like Teams Facilitator and Copilot in Outlook are taking more initiative themselves, which means you need to keep deciding what's a suggestion and what's a decision. And with systems like Gemini, you can see that access and permissions matter just as much as the convenience they offer.
Teams Facilitator now looks up answers itself during your meeting
From September, the Facilitator agent in Microsoft Teams can run a Copilot web search on its own. If a factual question in a meeting goes unanswered, the agent looks it up and posts the answer in the meeting chat, without anyone asking. You switch the feature on manually per meeting, and it's designed to respond less than once per meeting. You need Microsoft 365 Copilot Premium for it, so ask your admin whether your licence covers that. A button is also coming that lets participants report suspicious behaviour during or after a call, as part of a wider compliance approach via Purview. Notably, a related feature that let Copilot Studio agents join meetings themselves was quietly dropped from the roadmap in August. Turn the search feature on for the first time in an internal meeting, not a client call. The agent listens along with the transcript, spots an unanswered factual question, and looks it up on the web. The bar is set deliberately high, so it doesn't turn into a stream of interruptions. A web search draws on public sources, not on your own figures or agreements. So in a meeting about an internal budget or a supplier contract, the answer could sound confident and still be wrong. It does appear in the chat, though, and can later show up in the minutes as if it were a decision. So agree that an answer from the agent always counts as a suggestion, and should never go into the minutes unchecked.
Background for subscribersThis is a taster.
Subscribers read the whole Office management edition: every story, each with its background in plain language.
Subscribe →Not ready for that? Read the free letter first →
AI and tech, made simple.
This is the Office management edition. Subscribers get the background to every story, in plain language.
Subscribe
inklaretaal · Amsterdam · © 2026