The Contagion Tensor: A Framework for Measuring Output-Distribution Coupling in Multi-Agent LLM Systems -- and Auditing the Claims It Enables
arXiv:2606.28839v1 Announce Type: new Abstract: We introduce the Contagion Tensor, a measurement framework for quantifying how large language model (LLM) output distributions couple across modalities, agents, and time steps. From the tensor we derive the Coupling Amplification Factor (CAF), a family of ratio-based metrics sharing the form CAF = E[T_condition] / E[T_baseline], providing unitless, baseline-referenced measurement with bootstrap confidence intervals. We instantiate CAF in four variants and evaluate the strongest in a complete 2x2x2 block-orthogonal simulation design with modality-
The proliferation of multi-agent LLM systems necessitates robust measurement frameworks to understand their emergent behaviors and potential for amplified outputs.
This framework provides a critical tool for auditing the internal dynamics of complex AI systems, offering transparency and enabling safer, more predictable deployments.
The ability to quantify coupling and amplification within LLM systems shifts from qualitative observation to quantitative measurement, allowing for new levels of control and understanding.
- · AI safety researchers
- · Developers of multi-agent AI systems
- · Regulators overseeing AI deployments
- · Ethical AI auditors
- · Developers ignoring emergent system behaviors
- · Black-box AI system designers
- · AI systems prone to uncontrolled amplification
The Contagion Tensor allows for systematic identification of coupling amplification factors in multi-agent LLM systems.
This improved understanding leads to the development of more robust, controllable, and auditable AI agent architectures.
Enhanced trust and transparency in complex AI systems could accelerate their integration into critical domains, leading to widespread automation impacts.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG