Research

Paper

AI LLM March 19, 2026

Behavioral Fingerprints for LLM Endpoint Stability and Identity

Authors

Jonah Leshin, Manish Shah, Ian Timmis, Daniel Kang

Abstract

The consistency of AI-native applications depends on the behavioral consistency of the model endpoints that power them. Traditional reliability metrics such as uptime, latency and throughput do not capture behavioral change, and an endpoint can remain "healthy" while its effective model identity changes due to updates to weights, tokenizers, quantization, inference engines, kernels, caching, routing, or hardware. We introduce Stability Monitor, a black-box stability monitoring system that periodically fingerprints an endpoint by sampling outputs from a fixed prompt set and comparing the resulting output distributions over time. Fingerprints are compared using a summed energy distance statistic across prompts, with permutation-test p-values as evidence of distribution shift aggregated sequentially to detect change events and define stability periods. In controlled validation, Stability Monitor detects changes to model family, version, inference stack, quantization, and behavioral parameters. In real-world monitoring of the same model hosted by multiple providers, we observe substantial provider-to-provider and within-provider stability differences.

Metadata

arXiv ID: 2603.19022
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-03-19
Fetched: 2026-03-20 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.19022v1</id>\n    <title>Behavioral Fingerprints for LLM Endpoint Stability and Identity</title>\n    <updated>2026-03-19T15:23:23Z</updated>\n    <link href='https://arxiv.org/abs/2603.19022v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.19022v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>The consistency of AI-native applications depends on the behavioral consistency of the model endpoints that power them. Traditional reliability metrics such as uptime, latency and throughput do not capture behavioral change, and an endpoint can remain \"healthy\" while its effective model identity changes due to updates to weights, tokenizers, quantization, inference engines, kernels, caching, routing, or hardware. We introduce Stability Monitor, a black-box stability monitoring system that periodically fingerprints an endpoint by sampling outputs from a fixed prompt set and comparing the resulting output distributions over time. Fingerprints are compared using a summed energy distance statistic across prompts, with permutation-test p-values as evidence of distribution shift aggregated sequentially to detect change events and define stability periods. In controlled validation, Stability Monitor detects changes to model family, version, inference stack, quantization, and behavioral parameters. In real-world monitoring of the same model hosted by multiple providers, we observe substantial provider-to-provider and within-provider stability differences.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <published>2026-03-19T15:23:23Z</published>\n    <arxiv:comment>4 pages, 1 figure, submitted to CAIS 2026 System Demonstrations</arxiv:comment>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>Jonah Leshin</name>\n    </author>\n    <author>\n      <name>Manish Shah</name>\n    </author>\n    <author>\n      <name>Ian Timmis</name>\n    </author>\n    <author>\n      <name>Daniel Kang</name>\n    </author>\n  </entry>"
}