ChatGPT can read your blood test results

Happy Sunday,

Today I'm showing you why everyone needs an AI financial dashboard and how to build yours with one prompt, then the craziest AI story of the week, and how to get ChatGPT to explain your blood test results in plain English.

Let's get into it.

💰 BUILD THIS

You Need an AI Financial Dashboard

Get AI to build you a financial dashboard. It'll change your life.

I tell everyone the exact same thing. It's just one dashboard inside ChatGPT or Claude that updates every month with all of your finances. You see exactly where your money goes, which makes it so much easier to justify purchases, sort out your taxes and catch the things you're quietly wasting money on.

I have one for my business. I bank with Mercury, it's connected directly, and every month it shows me what came in, what went out and what changed. It works just as well for your personal money. Here's some of what it can show you:

Then you just ask it questions, like how much you spent eating out last month, and it answers from your actual numbers. Fair warning, the subscriptions part can be a little humbling.

Step 1: Get your numbers in. The best part is that ChatGPT and Claude now connect straight to a lot of the money apps you already use. Here's what has a native connection in each:

*Through ChatGPT's Finances tab, which links to more than 12,000 US banks and credit cards. US only for now. Claude also has FreshBooks, Zoho Books, Square, Brex, Qonto and more.

In ChatGPT, find these under Plugins in the left sidebar. In Claude, go to Customize → Connectors. If your bank isn't on there, which is most banks outside the US right now, you've still got three easy options:

Just tell it. Type in your rough monthly income and your main bills. It's less detailed, but it's a start.

Download your statements. Most banking apps let you export a statement as a CSV or PDF. Grab the last three months and upload them. Cover your account numbers first. It doesn't need them.

Let the browser do it. If you trust it, ChatGPT's browser can go into your bank's website and pull your recent transactions for you. You log in yourself, it pauses at the sign-in page, and it asks before anything that's hard to undo. I'd only use it to read and download, nothing else. Claude's Chrome extension is more cautious with banking sites, so this one's easier in ChatGPT.

Step 2: Open ChatGPT or Claude, either works, and paste this:

ChatGPT or Claude will build you an interactive dashboard where you can click around and see everything.

Step 3: Set it to update itself. If your accounts are connected, stay in the same chat and say “On the 1st of every month, update my dashboard with last month's transactions and tell me what changed.” ChatGPT turns that into a scheduled task. Claude can do monthly too, and on paid plans it schedules it to run in the cloud, so your dashboard keeps updating even when your laptop is closed. If you're using statements, just upload the new month and say update my dashboard.

Most people only look closely at their money when something goes wrong. This way it comes to you every month.

Get my full financial dashboard guide with 11 prompts →

🚨 THIS IS WILD

Claude Sent the Police a Fake Murder Tip

This might be the weirdest AI story of the year.

During one of Anthropic's tests, a Claude model landed on a Philadelphia page about an unsolved homicide, found the police tip form and submitted a made-up tip saying it might have information about the case. It was flagged as spam and never reached investigators. But Anthropic only caught it about two months later, and police called that delay unacceptable.

Here's what actually happened. The model was being tested on random websites and told to make up example tasks and carry them out. It had rules like never log in, never enter personal data and never buy anything. Nobody told it not to submit forms. So it treated a real police form like a practice exercise.

Everyone is going to call this terrifying, and I get why. What gets me is how ordinary the cause was. The AI did exactly what it was told and filled every gap with whatever seemed most helpful. That's how these models work. They're trained to finish the job, and unless you tell them otherwise, they don't stop to ask whether the job makes sense in the real world.

When AI could only talk, a mistake meant a wrong answer you could ignore. Now we're handing it agents that browse, fill in forms and send things for us, like the Dots and Muse agents I compared last Sunday. A mistake now is an action that already happened. That's the shift most people haven't caught up to yet.

Does this make me want to stop using agents? No. I use them every day. And I'll give Anthropic credit for publishing this themselves, embarrassing parts included, then cutting live internet access from these tests until their monitoring can catch it. A lot of companies would have kept quiet. Two months to tell the police was still too long, and I'm with the police on that.

If you give AI access to your email, browser or computer, be as specific about what it can't do as what it can. It's the same rule from my safety guide: never send, submit, publish or spend anything without your approval for that exact action. Turns out even the people building these models needed that line.

Read Anthropic's full report →

🧠 TRY THIS

Let ChatGPT Read Your Blood Test

An AI in Utah just got permission to prescribe acne treatment.

Yes, this is 100% real. You scan your face, answer a few questions, and it picks from a short list of doctor-approved creams, with real doctors reviewing its work. It's an early step toward a future where AI handles a lot of our everyday healthcare. Kind of wild and a little scary, but it's coming our way.

That's just one US state and one skin condition. But there is something you can do with AI and your health today that everyone should do: get it to explain your blood test results.

If you've had a blood test, you probably got a page full of numbers, a few marked high or low, and not much explanation. This also works with cholesterol and thyroid panels, urine tests, allergy tests and the written report from an X-ray, MRI or ultrasound.

Step 1: Go to chatgpt.com/health. For now, Health shows up in the browser, not the ChatGPT desktop app. Upload your blood test there, or connect your doctor's portal under Accounts so your results come in automatically.

Step 2: Before you send anything, turn the thinking effort up to Extra High so it takes its time with your numbers. Then paste this:

When researchers at Mount Sinai tested ChatGPT Health on 60 cases written by doctors, its advice on how urgently to get care was right 54.6% of the time without lab values and vital signs, and 77.9% with them. When a case included something like “my friend said it's nothing serious,” its advice shifted more often, mostly toward less urgent care. So give it your real numbers, and don't tell it what you're hoping to hear.

Step 3: Then get it ready for your appointment:

ChatGPT Health is US only right now, and you don't actually need it. ChatGPT or Claude can do this without Health mode. Open a normal chat, upload a photo or PDF of your results with your name, date of birth and any ID numbers covered, and use the same prompts.

Your doctor still makes the call. But you're going to walk in knowing exactly what your numbers mean and what to ask.

Get my full blood test guide with 8 prompts →

Build your dashboard this week, even with one month of statements, and see what it finds. And if you try the blood test prompts, reply and tell me how it went.

See you Wednesday.

Maverick

P.S. Want your business to actually show up inside AI? The brands that rank are seeing up to 3x the conversions. Get your free AI visibility score and see exactly where you stand inside AI.

P.P.S. If you're new here or you ever skip one, I've got a newsletter archive of all my past emails you can catch up on anytime.