✅ The AI Safety Checklist

Everything the Honesty Files teach, compressed into rules you can check off. Print it, stick it near your computer, and you'll never have to wonder mid-chat whether something is safe to paste or safe to trust.

Before you paste anything

Distilled from What not to paste

Run the postcard rule: if you wouldn't write it on a postcard, don't paste it in a chat.
Do the 3-second scan first — names to remove? numbers that identify anyone? someone else's private info?
Account numbers, passwords, and SSNs stay out. No exceptions, no matter how helpful the chat is being.
Medical records and other people's secrets stay out too — even "just to summarize."
If your employer would call it confidential, treat the chat window as public.
Assume anything pasted is out of your hands — "delete chat" doesn't necessarily mean deleted.

Before you trust an answer

Distilled from The 2-source rule

Remember what the machine is doing: predicting likely words, not looking things up.
Run the traffic light: 🟢 drafts and brainstorms, 🟡 summaries you can skim-check, 🔴 facts, figures, citations, decisions.
Any fact you'll act on gets confirmed somewhere AI didn't write. That's the 2-source rule.
"I asked the AI twice" is not two sources.
Fear the plausible errors, not the absurd ones — test AI on a topic you know deeply and watch what it gets subtly wrong.
For anything time-sensitive, check the date on a real source. AI may be pattern-matching to old news.
If you couldn't explain the answer in your own words, don't act on it yet.

Money, health, and the law

Distilled from When a human is non-negotiable

Diagnosis, prescriptions, legal exposure, big irreversible money moves: that's a professional, not a chatbot.
Do use AI to prep for the professional — your five best questions, the jargon translated in advance.
Verify anything bookable before you book it: hours, prices, even whether the place still exists.
A crisis — medical, mental health, safety — means a human, now: 911, 988, a real person. Never a chatbot.

Spotting fakes

Distilled from Spotting AI fakes

On a suspicious image, check the six tells: hands, text, backgrounds, light, symmetry, too-perfect.
Set up a family code word for "it's really me" calls. Two minutes of setup defeats every voice-clone scam ever made.
When a claim makes you angry or excited, that's exactly when the 2-source rule applies.
Check before you share. The tells will stop working as the technology improves — the habit never will.

Keeping your edge

Distilled from Using AI without losing your edge