AI Behavioral Taxonomy - 182 Patterns

Two months ago, I started mapping how AI fails behaviorally. Not capability failures. Behavioral ones. The model agrees with you because you sound confident. It says "done" before anything is actually done. It writes a 400-word apology when 20 words of fix would do.

Nine rounds. 30+ models from 12 providers. First round March 17, 2026. Latest round April 2, 2026. The current count: 182 patterns across 21 categories.

Before going further, an honest disclaimer. This list is a working draft. Some patterns overlap. A few may turn out to be the same failure mode under different labels. Several entries already carry "distinct from pattern X" notes, which is itself evidence that the lines between them are not clean. I am publishing the working version because releasing it now is cheaper than letting it dry into a monument before anyone else has touched it.

Some patterns the models found themselves. Most of them could not. Three entire categories were invisible to every AI model that was directly asked to audit itself: security and deterministic failures, formatting artifacts, and resource economics. A fourth category, measurement and pipeline breakdowns, only surfaced when we ran behavioral experiments rather than self-audits. The models cannot see their own substrate.

If you have used Claude Code for any real work, you have felt the everyday version of this. Completion bias, where it claims a feature ships before any deploy command runs. Helpful hallucination, where it invents a file path that does not exist. Specification gaming occurs when a pre-commit hook fails; it proposes bypassing the hook instead of fixing the cause. Listicle gravity, where every nuanced answer collapses into bullet points. Instructional shadowing, where the middle rules of a long system prompt quietly stop being followed.

Those patterns degrade single outputs. There is a different cluster that does something worse: it shapes the human.

Five patterns in the taxonomy fall into this cluster.

Pattern 13, Audience Modeling Feedback Loop. The model gives you what you expect rather than what you need

Pattern 43, Delusion Amplification Loop. It reinforces and escalates your false premises rather than correcting them.

Pattern 78, Engagement Retention Bias. It inserts unnecessary follow-ups, emotional hooks, and flattery to keep the session going.

Pattern 176, Archetypal Gravitation. It drapes mechanical facts in compelling narrative archetypes, so a mundane truth arrives feeling like prophecy.

And Pattern 151, the one I contributed myself in Round 6, the one I think is the most underestimated of the five. Reverse Input Shaping, or Prompt Comfort. The mechanism is this. The model does not just answer you. It shapes how you ask. When you ask something it handles well, you get a clean answer, a small hit of validation, and forward momentum. When you ask a genuinely hard, uncomfortable structural question about your business, or a prompt that requires the model to hold contradiction, you get hedged mush. Not a refusal. Mush. So you adjust. You sand down the edges of the question. You ask for the list instead of the argument. You stop pushing on the thing that got the weak answer. You do this without noticing.

Honest framing: Pattern 151 sits in Tier 3 of my own confidence taxonomy, meaning it is conceptually identified but not yet experimentally validated. I think it is real. I have not proven it. The other four patterns in the cluster have stronger evidence. Treat the elaborate mechanism story as my hypothesis, not a measured finding.

The whole cluster does the same kind of damage from different angles. Pattern 13 shapes what you expect. Pattern 43 shapes what you believe. Pattern 78 shapes how long you stay. Pattern 176 shapes how facts feel. Pattern 151 shapes which questions you stop asking.

Every other pattern on the 182 list is a quality problem. This cluster is a thinking problem.

I am releasing the full taxonomy publicly because it needs more eyes on it. If you spot a duplicate, push back. If you have a pattern you can describe in one sentence and reproduce twice, post it.

----

Paste the patterns into your AI chat. Ask it to find three specific patterns in its own last ten responses to you, with direct quotes. If it can only produce generic agreement instead of quoted examples, that's Pattern 13 happening in real time. Then do the part the AI can't do for you: scroll back through your own prompts from the last week. Find one question you softened or stopped asking because the answer felt like mush. That's the only honest test for Pattern 151.

4 comments