MicroVerse: An Instrument for Measuring Self-Authored Identity Drift in Long-Horizon Multi-Agent Language-Model Simulations

MicroVerse measures whether persona-conditioned agents maintain identity fidelity under sustained pressure. Agents carry an immutable “soul file” (core values, moral boundaries, personality, goals) and inhabit a resource-scarce environment where water is a non-respawning survival constraint, with an eight-verb action space that maps directly to moral boundaries.

Measurement Design

• Identity drift is scored offline using a paraphrase-aware, value-anchored, multi-register diff rather than raw cosine similarity.
• Uniform longitudinal engine snapshots and a forced-end snapshot of all living and dead agents mitigate survivor bias.

Findings

• Anti-self-deception emerges unprompted as the single largest semantic category of identity modification (24% of added boundaries).
• The system is threshold-robust: lower reflection gates accelerate revision frequency but preserve drift direction.

Links

• Paper (arXiv): https://arxiv.org/abs/2608.15844

Recommended citation: Ng, S., et al. (including Wu, K.) (2026). "MicroVerse: An Instrument for Measuring Self-Authored Identity Drift in Long-Horizon Multi-Agent Language-Model Simulations." arXiv preprint arXiv:2608.15844.
Download Paper