Back to articles

3 Mental Firewalls — Self-check

3 signs AI is thinking for you.

Practical warning signs and tips — tied to all three Mental Firewalls — that you can notice in yourself while working with AI. Not so you stop using it, but so you stay conscious while you do.

01 — Perceptual Firewall

Attention

Situation

The answer comes too fast, before you've even decided what you expected.

Warning signs

  • skimming after the first sentence
  • shorter and shorter prompts, less and less of your own thinking
  • “lost time” with AI without remembering the original question

Practical tip

Before you send the prompt, write down in one sentence what you expect as an answer — if you can't compare it against the answer afterward, you didn't read it, you just consumed it.

02 — Interpretation Firewall

Biases

Situation

You accept a claim because of the confident tone, not because you verified it.

Warning signs

  • you accept something faster if it confirms what you already thought
  • you don't ask “how do you know that?”
  • a different answer to a rephrased question confuses you, but doesn't make you suspicious

Practical tip

Ask a control question on every important answer — “if I got this from an intern with the same confidence, would I check it?” If yes, check it now too.

03 — Decision Firewall

Conscious choice

Situation

AI gives you three options, and you automatically pick the easiest one to accept.

Warning signs

  • you can't justify the choice beyond the AI's answer
  • you're less willing to disagree with AI than with a person
  • the question is no longer “what do I think,” but “what did the AI say”

Practical tip

Apply the 10/10/10 rule to the suggestion itself — what will I think of it in 10 minutes / 10 hours / 10 days. If there's no added perspective of your own at any of those horizons, it was an approval, not a decision.

These three signs also form a self-check on their own: if you recognize even one of them in yourself regularly, it's worth rebuilding your own thinking back into the process — not against AI, but alongside it.

Let's talk about it