Hi there,
The firms whose whole job is warning everyone about AI getting things wrong just got caught doing it themselves.
This week KPMG, of all people, published a report that a detection company found was littered with AI hallucinations. Confident, invented details, dropped into a professional document by a firm that has spent years telling clients to watch for exactly that. If it can happen to them, it can happen to you.
The good news is the fix is a habit, not a tool. So this week, how to stop AI quietly embarrassing you.
This week in AI
The short version. KPMG’s stumble wasn’t a one-off. A new survey now puts hallucinations, the confident false answers AI hands you, above job losses as the number one worry among people who use these tools. More than six in ten call it their biggest barrier to using AI properly. The takeaway isn’t stop using it. It’s never let its work leave the building unchecked. Which is exactly this week’s job.
The practical bit
The more confident it sounds, the harder you look.
Here’s the thing about AI mistakes. They don’t look like mistakes. A hallucination comes back polished, specific and sure of itself, in the same tone the AI uses when it’s right. No wobble, no “I think”. That’s exactly what slips it past a careful reader.
So before anything AI touches leaves the building, run a two-minute check. It comes down to knowing where it invents.
Usually fine to trust
Shapeless work. Tidy this up, summarise the gist, give me a rough first draft I’ll rewrite. If it’s a bit off you’ll spot it, no harm done.
Always check
Anything specific. Names, numbers, dates, quotes, statistics and above all sources. If it cites a study, a law or a link, assume it could be invented until you have seen the real thing. This is where AI makes things up and where it costs you.
One more tell. The more certain the answer sounds, the harder you look. Confidence is not evidence. Spend two minutes on the sharp bits before you hit send. That keeps you out of the KPMG story.
Free resources
Spotting when AI makes things up
A free lesson from our Prompting 101 course that goes further than a two-minute check. It walks through what today’s AI is genuinely bad at and how to catch it inventing. If this issue landed, it’s your next fifteen minutes well spent.
New here? You’ll set up a free community login on the way in.
What’s on
The sessions below ran in July 2026. For what is coming up, see upcoming events.
This Thursday. Last chance to join us in person. In one morning you walk out with your AI strategy and policy written by you. That’s the Build Along, on 9 July. There’s a free afternoon session too, so come for one or both.
Build Along with AI
9 July 2026 · White City, London · 11:00 to 13:00
Walk in unsure about AI. Walk out with an AI strategy and a one-page policy you wrote yourself, in one morning. A working session with us in the room, not a talk.
Free virtual waitlist
The AI Foundry Live
9 July 2026 · White City, London · 13:30 to 16:30
Our quarterly free live event. Two short talks from guest speakers, then time to compare notes with other UK business owners over an afternoon coffee.
Speaking this time: Mahdi Shariff from Humble AI and Hannah Morris from Let’s Trailblaze.
Free, uncapped
When you’re ready to go further than one job, this is the room to do it in, with us there to help. Bring the person who really owns this in your business.
More sessions
The sessions below ran in July 2026. For what is coming up, see upcoming events.
More coming up this month
Can’t make the 9th or want to go further? Three more sessions through July, all free and open to anyone.
AI does the work. You do the checking. That’s the deal.
AI is genuinely good now. It will draft, summarise and research faster than you can. And it does all of that with total confidence, whether it’s right or wrong. That last part is the catch. It’s the whole reason you stay in the loop. Let it do the work. You do the checking. Nobody has ever regretted the two minutes.
Next week, how to get the check out of your own head and into your team, so it doesn’t all depend on you.
Hit reply and tell me the best AI mistake you’ve caught before it went out. I read every reply.
Know someone who trusts AI a little too much? Forward this on.
Understand first. Then act.
Richard