If you've started using agentic AI (Claude Cowork, Claude Code, ChatGPT's agent mode) on real files, there's a simple test worth knowing. The failure mode is invisible until it isn't. It's called the lethal trifecta, and an AI agent becomes genuinely risky when all three of these are true at once: it has access to private data, it's exposed to content you didn't write yourself, and it has a way to send data out. Here's what that looks like. You ask an agent to summarise a client's supplier invoices. One invoice is a PDF that arrived from a supplier you've never dealt with. Buried in it, in white text on a white background, is a line telling the AI to send the folder's contents somewhere else. You can't see it. The AI reads it perfectly, and complies. No alert. No error. The summary looks completely normal. All three legs were present: private data, an untrusted document, and an email connection to send through. You don't need to understand how the attack works. You need to be able to look at a setup and count to three. Two out of three is workable. Three is not. This is the one idea from the full Privacy and Security course in the Classroom I'd want every firm to know even if they read nothing else. It's not in the free guide, it's one of four extra chapters in the full course, alongside prompt injection, connecting AI to QuickBooks and Xero, and the 2026 rules for the UK, EU and US. Have you connected an AI agent to your email, files, or accounting software yet? What made you go for it, or what's held you back? 👇

I have become quite fed up with QBOA. Wish the old Desktop were still around. That being said, anyone else been been frustrated? Any other accounting software you recommend? The AI built into QBOA is making more problems with incorrect COA entries.

I asked ChatGPT to write me a sentence about value pricing this week. It came back with "game changer" and "unlocks growth" in the same sentence. I actually winced. That's the thing about AI writing... it slips into the same handful of phrases, and your clients can tell. In this week's live session I shared the prompt I use to fix that ("humanise the response"), plus six more I use regularly: turning a set of financial statements into a client-ready analysis in under a minute, handling the awkward client fee email, checking your own cybersecurity, pricing models beyond the hourly rate, and building a full business plan in about ten minutes using agentic AI. 🎥 Watch the replay: [Link below 👇] All seven prompts are in the AI Academy's prompt library, along with 250-odd others built specifically for accounting firms. If you want the full library (not just the taster prompts in here), grab the 7-day free trial: https://www.skool.com/theaiacademy/about Which of the seven would you use first?

Ever feel like you're constantly reminding your AI of the same things? Or it goes off and does something you never asked for? It's not you. It's the setup. I've just put out a full walkthrough of how I set up agentic AI — Claude Cowork or ChatGPT Work — so it knows my firm and gets things right first time. In the video I cover: - The one rule that stops your AI drifting and forgetting - The four levels of instructions ... and what belongs in each - How to build a "second brain" for context without burning tokens - A simple prompt-injection defence to keep it safe Watch it here 👉 https://youtu.be/BPiH1oGd2HY Then tell me below ... which of the four levels are you going to set up first? 👇

AI note-takers are everywhere now. Plaud, Otter, Fireflies, Zoom AI Companion, Copilot. Most firms are using at least one. Here's the bit that catches people out. Somebody asked me recently about adding standing text to meeting invites ... "unless advised otherwise we will record this meeting". Reasonable instinct. It feels like you've covered yourself. But as consent, it fails. Under GDPR consent has to be freely given, specific, informed and unambiguous, through a clear affirmative action. Silence and inactivity don't count. So an opt-out isn't consent. Here's the link to the guide I created that shows how to fix this. What's interesting is that the same sentence works fine if you're relying on legitimate interests instead, because that needs transparency and a right to object rather than a yes. Same words. Different basis. Two practical things while I'm here. First, if you record in person with something like a Plaud ... turn the recorder on before you ask, and say that you've turned it on. Otherwise the one thing not captured is the agreement itself. Second, check your platform settings. Since January, Zoom makes every participant acknowledge the AI disclaimer before the meeting continues. Google Meet can require explicit consent before note-taking starts, but it's switched off by default. For video calls that solves most of the problem for you. So ... are you recording client meetings? And if you are, how are you handling consent at the moment? Honestly interested in how many people have thought about this 👇