Book 2 · Chapter 9: Advanced Safety and Judgment
This page gives you a three-source verification method and a printable Book 2 safety checklist.
This page assumes you've read the chapter. It hands you the tools, not the lessons.
Three-Source Verification Card
When an AI makes a factual claim that matters to your work, you verify it against three independent sources. Not three mentions of the same fact (that doesn't count). Three different sources, from different organizations, different methodologies, different perspectives, that each independently confirm the same information.
What makes sources independent
Three sources that all cite the same original report aren't truly independent. If Glassdoor, LinkedIn, and a blog post all reference the same Bureau of Labor Statistics dataset, that's one source with three amplifiers. Look for sources that arrived at their conclusions independently: different methodologies, different data sets, different research teams.
What counts as a source
- A Reddit comment is not a source.
- An AI's citation of a study you haven't read is not a source.
- Direct evidence you can read yourself is a source.
Use three-source verification for
- Research-based decisions.
- You're building a proposal around market data, competitor information, or industry trends. Do the sources exist? Do they say what the AI said they said?
- Unusual claims.
- If the AI tells you something that surprises you or contradicts what you thought you knew, that's a verification trigger. Your surprise is a signal that you should check.
- Money and contracts.
- If the AI's claim impacts your financial decision, your liability, or a legal agreement, three sources isn't overkill. It's the baseline.
- Delegation downstream.
- If you're about to hand this work to someone else (a colleague, a contractor, a client), you need to be confident enough to stand behind it. That means verification that goes beyond "this seems reasonable."
From the Three-Source Verification section.
The Book 2 Safety Checklist
Print this. Put it somewhere you'll see it. Go through it weekly for the next month until it becomes automatic.
Before you share anything with AI
- Does this contain information I wouldn't want public?
- Is there a way to ask this question without sharing sensitive details?
- Am I comfortable with this information being in my AI's context indefinitely?
- Would sharing this violate my company's data policy?
When AI produces research or fact-based claims
- Have I verified the most important claims against independent sources?
- For critical claims, have I found three truly independent sources?
- Do the sources say what the AI said they said?
- Is the information current, or is the AI relying on outdated data?
- If I'm making a financial or medical decision on this, have I done 3-source verification?
When AI produces work that you'll hand off
- Have I read this closely, or just skimmed it?
- Do the conclusions make sense, or am I just trusting because it's from AI?
- What would be wrong with this output? Have I checked for those specific things?
- Could someone else use this work without realizing something's missing or wrong?
- Am I willing to attach my name to this? If not, why not?
When you're building multi-step workflows
- Which outputs are most critical to get right?
- Have I built verification gates into those steps?
- Does each step's output get used correctly in the next step?
- Am I catching errors at the point where they're introduced, or downstream?
- Have I red-teamed this workflow to find where it could break?
On your own skill and judgment
- Have I done a version of this work myself recently?
- Do I understand the problem well enough to catch when AI gets it wrong?
- Am I delegating work, or delegating thinking?
- If the AI broke tomorrow, could I still do my job?
- Could I explain to someone else how this work gets done?
On your AI usage broadly
- Am I using AI to move faster on things I understand, or to avoid understanding things?
- Have I built habits that would break if the AI goes down?
- Am I relying on AI for something that's core to my expertise?
- Have I stayed aligned with my company's policies on AI and data?
- Am I staying sharp on the skills I'm offloading?
From the Book 2 Safety Checklist section.