Identity as Attractor: Geometric Evidence for Persistent Agent Architecture in LLM Activation Space

TL;DR

This paper demonstrates that the internal representations of an agent's identity in large language models form attractor-like structures, with semantic content playing a key role, supported by experiments across models and conditions.

Contribution

It provides the first empirical evidence that agent identity documents induce attractor-like geometry in LLM activation space, emphasizing semantic influence.

Findings

Paraphrases converge to a tighter cluster than controls, indicating attractor-like behavior.

Cross-architecture experiments confirm the generality of the effect.

Reading about the agent shifts internal states toward the attractor, showing knowledge of identity influences representations.

Abstract

Large language models map semantically related prompts to similar internal representations -- a phenomenon interpretable as attractor-like dynamics. We ask whether the identity document of a persistent cognitive agent (its cognitive_core) exhibits analogous attractor-like behavior. We present a controlled experiment on Llama 3.1 8B Instruct, comparing hidden states of an original cognitive_core (Condition A), seven paraphrases (Condition B), and seven structurally matched controls (Condition C). Mean-pooled states at layers 8, 16, and 24 show that paraphrases converge to a tighter cluster than controls (Cohen's d > 1.88, p < 10^{-27}, Bonferroni-corrected). Replication on Gemma 2 9B confirms cross-architecture generalizability. Ablations suggest the effect is primarily semantic rather than structural, and that structural completeness appears necessary to reach the attractor region. An exploratory experiment shows that reading a scientific description of the agent shifts internal state toward the attractor -- closer than a sham preprint -- distinguishing knowing about an identity from operating as that identity. These results provide representational evidence that agent identity documents induce attractor-like geometry in LLM activation space.