The Sentinel Code 🐉
Seven rules I keep reminding myself of. And that everyone who works a lot with AI should know.
1
Well written doesn't mean right.
AI always writes beautifully. Even when it's wrong.
2
Confident doesn't mean correct.
AI sounds just as sure when it's wrong as when it's right.
3
Think first, then ask.
If you always ask the AI first, you forget how to think it through yourself.
4
Distrust isn't enough. Checking is what protects you.
People who don't trust AI still fall for it.
5
Check the most important thing yourself.
Not everything. Just the one thing that really matters. Ideally at the source.
6
If the AI agrees with you, look twice.
AI tends to agree with us. And the mistakes we like are the ones we miss most.
7
The more important it is, the more carefully you check.
With health, money and the law, mistakes are expensive.
And now you're probably thinking: That won't happen to me.
In one study, up to 80 percent missed the errors. Even though they knew the subject.
These rules are based on current research.
Sources
- Rules 1 and 2 and the closing line: de Sá Siqueira, Bosse, Müller (Radboud University), Computers in Human Behavior: Artificial Humans, 2026. doi.org/10.1016/j.chbah.2026.100412
- Rule 3: Rani, Danry, Liang, Lippman, Maes (MIT Media Lab), CHI 2026. arxiv.org/abs/2510.01537
- Rules 4 and 7: Goulas, Megalokonomou, Sotirakopoulos, PNAS Nexus, 2026. doi.org/10.1093/pnasnexus/pgag146
- Rule 5: Hruschka, Appel, PLOS ONE, 2023. doi.org/10.1371/journal.pone.0283238
- Rule 5: Manhede et al., AI & SOCIETY, 2026. doi.org/10.1007/s00146-026-03373-3
- Rule 6: Cheng, Jurafsky et al., Science, 2026. doi.org/10.1126/science.aec8352