Your Personal AI Insurance Policy

Friday we wrote about the lawyer who just lost his license for trusting ChatGPT. Today we write about the system he didn't have.

Since 2023, the Charlotin tracker has documented 1,300+ court cases where someone used AI output containing fabricated facts — citations, statutes, quotes, precedents. Roughly 800 came from US courts. The largest sanction so far: $109,700, in Oregon.

The cases aren't all lawyers. There's a marketing manager who briefed a CEO using an invented industry benchmark. A financial advisor who quoted a regulation that doesn't exist. An analyst whose Q3 deck was anchored to a fabricated competitor stat. Most never make the news. They show up as quiet firings, lost accounts, and stalled careers.

What every case has in common: confident AI output → trusted user → no verification → consequences. The chain is identical. Only the price tag changes.

AI Doesn't Lie. It Lies Confidently.

That's the part that makes it dangerous.

AI doesn't say "I'm guessing." It states facts in the same calm, polished tone whether the fact is real or invented. There is no detectable difference between "the Federal Reserve raised rates by 25 basis points last quarter" (which you can verify in 4 seconds) and "Section 412(b)(2) of the Securities Act prohibits…" (which may be entirely fictional).

Your brain registers both as authoritative because they sound identical. That's not a bug AI vendors will fix soon — it's structural to how language models work. They're built to sound right, not to be right. Even the best frontier models hallucinate on a meaningful percentage of factual queries. The rate gets lower every release. It will never reach zero.

Which means the responsibility for catching hallucinations falls entirely on the human pressing send. Every time. Forever.

Why "Just Check It" Fails Almost Everyone

The default advice — "always verify AI output" — fails for the same reason every generic productivity tip fails: it ignores your context.

A lawyer running a citation audit needs 15+ minutes per brief. A copywriter pitching client subject lines has 90 seconds. A regulator-facing financial advisor needs a totally different checklist than an internal analyst writing a status update. Telling all four of them the same thing — "verify your AI" — guarantees three of them won't.

Generic advice gives you none of those. That's why most professionals don't actually verify — the advice they've been handed is either too heavy to follow or too vague to apply. So they ship. And every time they get away with it, the bar for the next time drops a little lower. Until it doesn't.

The Fix: A Protocol Built for You

The prompt below interviews you in five questions — work type, audience, worst-case scenario, top fear, time budget — then designs a verification protocol calibrated to your exact situation.

It produces:

• A 3-step verification protocol matched to your time budget
• The specific questions to ask AI before trusting any output
• A red-flag checklist tuned to your failure mode
• One concrete example applied to your real workflow

Run it once. Save the output. Use it on the next AI draft you send.

Below is what the same prompt produced for four very different professionals — same five questions, wildly different protocols.

The Bottom Line

The lawyer didn't get sanctioned because AI made things up. AI was always going to make things up.

He got sanctioned because he didn't have a system.

Generic advice doesn't give you a system. The prompt above does.

Run it. Save the output. Then forward this to the colleague most likely to be next.