🔒 Machines That Talk · Day 19

The Privacy Shredder

Whatever you paste into a chatbot leaves your device — it can be stored, logged, even read by a human later. So before you hit send, shred the secrets you don't need to share. Let's hunt them down.

↓ start shredding
🛡️ Lab · The Privacy Scanner

Find the secrets. Shred them.

Here's a real message a kid might send an AI. The scanner below auto-highlights anything sensitive. Click each glowing bit to replace it with [REDACTED] — or hit the big shred button. Watch the risk meter fall and the token count shrink.

0secrets still exposed
0tokens being sent
🛡️ Privacy Risk: —
The safe version still does the job: — tokens — same help, zero secrets leaked.

🔵 email   🟣 phone   🔴 API key   🟡 address. The AI only needs to know it's about volcanoes — not where you live!

🐞 Break-It Challenge

See what the server saw

Redacting on your screen feels safe... but the very first version already left your laptop. Flip the switch to see the raw log a company's server might store — secrets and all.

log hidden
🏷️ The pro wordSending only what the AI actually needs is called data minimization. The private stuff — your name, address, phone, email — is your PII (personally identifiable information). Blacking it out before it leaves your device is redaction. Pros do all three, every single time.
🎯 Boss Build · Redact Patrol

Spot every secret

Three messages, one job: click every sensitive bit — and nothing harmless. Miss one and it leaks; over-shred and the message stops making sense. Get 3 clean sweeps to earn the final badge!

Message 1 of 3 — click the secrets:
Clean: 0 / 3
🚀 Beyond the Basics

How the pros really do it

"Just be careful" is a start — but real companies build privacy in so nobody has to remember. Here's how it grows up:

Just be careful (what you'd try first)

Eyeball every message and delete secrets by hand. ✅ Free, no setup. ⚠️ Humans forget, get lazy, or miss things — one slip and a secret is gone.

🤖

Auto-redact before send (what you built)

A scanner strips out PII automatically before the prompt leaves. ✅ Catches the obvious stuff every time. ⚠️ Regex misses clever or misspelled secrets.

🏢

Enterprise controls (the real deal)

Zero-retention & no-training modes, on-device local models, encryption, and audit logs. ✅ Safest by far. ⚠️ Costs money and takes real setup.

🏷️ Pro names to look upThe grown-up toolkit is PII redaction, data minimization, zero-retention modes, local inference (the model runs on your own machine), and the privacy law GDPR.

🎉 Day 19 Complete!

You learned the golden rule of talking to AI: whatever you send leaves your device, so send only what the job needs. You can now spot and shred PII — email, phone, address, API keys — before you ever hit that button.

Day 20: Build Your BudgetBot →
Tomorrow: you put it ALL together and build your own BudgetBot — a chatbot that watches its tokens, its cost, and its secrets. 🤖💰
AI Adventure · Machines That Talk · Day 19 — made for young inventors 🚀