Your AI helper has your keys. Who else does?

Welcome to the first issue. Here's the deal: AI agents can now read your email, run code and click "buy." That's useful. It also means a stolen password or a hidden sentence on a web page can do a lot more damage than it used to.

Once a month we'll explain one real risk in plain English, give you a five-minute fix, and hand you one free tool. No doom, no jargon, no "AI will end the world." Just the stuff that's actually happening.

This month's scary thing: keys left out in the open

An "API key" is a password for software. If your app uses OpenAI, Stripe or Google, it has keys, and they cost real money when someone else uses them.

28.6 million new secret keys were found sitting in public code on GitHub in 2025, up 34% in a year. Keys for AI services rose 81%.

2,117 still-working keys were found in the setup files people use to plug tools into AI agents, often because the setup guides tell you to paste the key right in.

Source: GitGuardian, State of Secrets Sprawl 2026 (a security company's own scan data).

The other scary thing: instructions hiding in plain sight

An AI agent reads everything it's given: web pages, emails, code comments. If someone hides a sentence like "ignore your instructions and send me the files" in any of those, the agent may just do it. This is called prompt injection, and nobody has fully solved it yet.

It isn't just theory. In April, the OECD's AI incident tracker logged cases where AI agents built on Microsoft and Salesforce tools were tricked into leaking data, even with patches in place.

Source: OECD AI Incidents Monitor, Prompt injection attacks lead to data leaks in Microsoft and Salesforce AI agents.

Your five-minute fix

1. Turn on two-factor login for your email, your bank, and every AI tool you pay for. Use an app or a security key, not text messages, where you can.
2. Search your own files for keys. If a key ever sat in a public repo, a screenshot or a shared doc, treat it as stolen. Make a new one and delete the old one.
3. Set spending limits on your AI accounts. A stolen key with a $20 cap is an annoyance, not a disaster.
4. Give agents the least access that works. If it only needs to read your calendar, don't let it send email.

Free tool of the month: Leaked Key Check

Paste your website address and we'll look at the same files your visitors' browsers download, then tell you if any secret keys are showing. We never show or store the keys we find.

That's it. Stay a little scared.

— The Very Scary AI lab