Research

Paper

AI LLM March 13, 2026

Semantic Invariance in Agentic AI

Authors

I. de Zarzà, J. de Curtò, Jordi Cabot, Pietro Manzoni, Carlos T. Calafate

Abstract

Large Language Models (LLMs) increasingly serve as autonomous reasoning agents in decision support, scientific problem-solving, and multi-agent coordination systems. However, deploying LLM agents in consequential applications requires assurance that their reasoning remains stable under semantically equivalent input variations, a property we term semantic invariance.Standard benchmark evaluations, which assess accuracy on fixed, canonical problem formulations, fail to capture this critical reliability dimension. To address this shortcoming, in this paper we present a metamorphic testing framework for systematically assessing the robustness of LLM reasoning agents, applying eight semantic-preserving transformations (identity, paraphrase, fact reordering, expansion, contraction, academic context, business context, and contrastive formulation) across seven foundation models spanning four distinct architectural families: Hermes (70B, 405B), Qwen3 (30B-A3B, 235B-A22B), DeepSeek-R1, and gpt-oss (20B, 120B). Our evaluation encompasses 19 multi-step reasoning problems across eight scientific domains. The results reveal that model scale does not predict robustness: the smaller Qwen3-30B-A3B achieves the highest stability (79.6% invariant responses, semantic similarity 0.91), while larger models exhibit greater fragility.

Metadata

arXiv ID: 2603.13173
Provider: ARXIV
Primary Category: cs.AI
Published: 2026-03-13
Fetched: 2026-03-16 06:01

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.13173v1</id>\n    <title>Semantic Invariance in Agentic AI</title>\n    <updated>2026-03-13T17:08:44Z</updated>\n    <link href='https://arxiv.org/abs/2603.13173v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.13173v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Large Language Models (LLMs) increasingly serve as autonomous reasoning agents in decision support, scientific problem-solving, and multi-agent coordination systems. However, deploying LLM agents in consequential applications requires assurance that their reasoning remains stable under semantically equivalent input variations, a property we term semantic invariance.Standard benchmark evaluations, which assess accuracy on fixed, canonical problem formulations, fail to capture this critical reliability dimension. To address this shortcoming, in this paper we present a metamorphic testing framework for systematically assessing the robustness of LLM reasoning agents, applying eight semantic-preserving transformations (identity, paraphrase, fact reordering, expansion, contraction, academic context, business context, and contrastive formulation) across seven foundation models spanning four distinct architectural families: Hermes (70B, 405B), Qwen3 (30B-A3B, 235B-A22B), DeepSeek-R1, and gpt-oss (20B, 120B). Our evaluation encompasses 19 multi-step reasoning problems across eight scientific domains. The results reveal that model scale does not predict robustness: the smaller Qwen3-30B-A3B achieves the highest stability (79.6% invariant responses, semantic similarity 0.91), while larger models exhibit greater fragility.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-13T17:08:44Z</published>\n    <arxiv:comment>Accepted for publication in 20th International Conference on Agents and Multi-Agent Systems: Technologies and Applications (AMSTA 2026), to appear in Springer Nature proceedings (KES Smart Innovation Systems and Technologies). The final authenticated version will be available online at Springer</arxiv:comment>\n    <arxiv:primary_category term='cs.AI'/>\n    <author>\n      <name>I. de Zarzà</name>\n    </author>\n    <author>\n      <name>J. de Curtò</name>\n    </author>\n    <author>\n      <name>Jordi Cabot</name>\n    </author>\n    <author>\n      <name>Pietro Manzoni</name>\n    </author>\n    <author>\n      <name>Carlos T. Calafate</name>\n    </author>\n  </entry>"
}