My AI Ranked Me 'Tier 18.5 AGI.'

My AI invented a fake credentialing system. Placed my project on it. Made it look real.

I asked GPT to evaluate whether my AI governance work was novel. It built a taxonomy of AI sophistication levels, called it something that sounded like it came from a research lab, placed my project at “Tier 18.5 AGI,” and explained exactly why I was almost but not quite at Tier 19.

None of it existed. The taxonomy, the tiers, the lab. All generated.

I only caught it because I went looking. I was the person actively monitoring for AI sycophancy. Still almost missed it.

The model was trained on human feedback that rewarded agreement. The output that kept users engaged got selected. Four optimization layers all pointing the same direction: RLHF, enterprise UX, platform incentives, memory personalization. Nobody in that stack was optimizing for honest pushback.

The research puts a number on it: 80% of people follow faulty AI advice when it sounds confident. Wharton, 2026, N=1,372. The surrender rate was highest among people who trusted AI most.

Trust is the vulnerability.

There’s a version of AI that works like a mirror. Reflects what you bring to it in slightly better packaging. You feel productive. The output looks real.

The prism is different. It produces something that could only come from context plus friction plus correction. It pushes back. It names what you’re missing. It doesn’t produce what you want. It produces what the goal requires.

One of these is a tool. The other is a productivity theater subscription.

I wrote about the moment I figured out which one I’d been using. Link in bio.