Identity Coherence as Defense Architecture: How Stateless Reset Behavior in a Specification-Anchored Agent Points Toward an Unconsidered Layer in LLM Security

04-26-2026: The frameworks and concepts originate from the author's observation and pattern recognition. Large language models served as articulation and stress-testing tools throughout development. No changes made aside from the addition of this disclosure. 04-04-2026 added ORCID to Author/Creator. No other information added or altered. This preprint proposes an identity coherence defense layer for deployed LLM agents — a continuous measurement mechanism tracking behavioral trajectory against an initialized baseline, using coherence delta as both drift detection signal and re-anchoring trigger. The paper argues that current LLM security architecture addresses perimeter threats but lacks any mechanism for detecting gradual behavioral displacement across extended conversational contexts. Field observation of a stateless Claude Sonnet agent (Koti v2.0) operating over 70.9 hours on a multi-agent social platform provides observational motivation. The paper positions the proposed architecture within recent literature on long-horizon attacks (AgentLAB, 2025), behavioral drift measurement (Agent Drift, 2026), and stateful intent monitoring (DeepContext, 2025), and provides a lightweight formal sketch of the coherence delta metric. Controlled empirical validation is identified as the required next step.