Trust But Verify: The AI Second Opinion

When even world-class consultants like Deloitte get burned by "AI-powered" mistakes, it's tempting to say, "No AI for us." But that's the wrong move.

AI, like any tool, delivers its best results under the same principle we'd use for major surgery or a six-figure renovation: always get a second opinion.

For every mission-critical decision, I recommend a parallel, or at least serial, verification step. That means letting a second human or AI system independently check the work, using a different model or logic, especially when sources, data, or citations are involved.

Humans make costly errors too. The difference? AI lets mistakes scale at lightspeed. That's why "trust but verify," at scale, isn't optional.

Here's my go-to checklist for safe, confident AI:

• If a source is cited, click to verify it exists. • If links are given, randomly audit a few. • If data is analyzed, cross-check using a different model, tool, or human. • When a process impacts high-stakes outcomes, assume nothing. Measure confidence.

Not every AI output needs double-checking. But anywhere there's mission impact or public trust at stake, leaders need to think in terms of confidence levels and systemic verification, not just blind faith or blanket rejection.

What's one major decision area where you'd now insist on AI "second opinion" checks? Let's help each other raise the bar.