# Multi-Agent LLM Systems

4 related articles

Subscribe to updates

arXiv Oct 7, 2026
TGCC uses cross-layer trust gating so a compromised LLM agent self-revokes within steps instead of escalating
The work turns a descriptive layered trust model into an operational controller: a cross-layer synergy operator propagates prerequisite-layer deficits into dependent layers, a no-regret online estimator grounds per-layer weights in observed failures, and Trust-Gated Capability Control issues short-lived, revocable grants only when composite trust and the relevant prerequisite layers clear capability-specific thresholds; the authors prove and numerically confirm that a stealthy compromise inflating behavioral trust while degrading a prerequisite layer cannot escalate privilege and instead self-revokes within a few interactions.
The work turns a descriptive layered trust model into an operational controller: a cross-layer synergy operator propagates prerequisite-layer deficits into dependent layers, a no-regret online estimator grounds per-layer weights in observed failures, and Trust-Gated Capability Control issues short-lived, revocable grants only when composite trust and the relevant prerequisite layers clear capability-specific thresholds; the authors prove and numerically confirm that a stealthy compromise inflating behavioral trust while degrading a prerequisite layer cannot escalate privilege and instead self-revokes within a few interactions.
BRAC University Rajshahi University of Engineering and Technology Ahsanullah University of Science and Technology
Mehedi hasan Nipu（First author）
Trust-Gated Capability Control Multi-Agent LLM Systems Agent Safety
arXiv Sep 27, 2026
Across 50 stateless LLM agents updating predictions from locally visible peers, the study identifies synchronised, twisted, and chimera-like collective regimes, finding reasoning effort and communication topology govern different aspects of coordination
Studying N = 50 stateless LLM agents that update predictions only from locally visible peers and characterizing behavior with both global and local measures of agreement, the work identifies three collective regimes—synchronised, twisted (locally ordered but globally incoherent), and chimera-like (coherent and incoherent subpopulations coexisting); increasing reasoning effort in gpt-5-mini shifts panels from variable, often fragmented outcomes toward locally ordered twisted states, while increasing communication connectivity drives them toward global synchronisation, and low spatial heterogeneity does not guarantee global consensus—40% of trials with Delta Z below 0.03 retain a twisted configuration through the final 20 turns.
Studying N = 50 stateless LLM agents that update predictions only from locally visible peers and characterizing behavior with both global and local measures of agreement, the work identifies three collective regimes—synchronised, twisted (locally ordered but globally incoherent), and chimera-like (coherent and incoherent subpopulations coexisting); increasing reasoning effort in gpt-5-mini shifts panels from variable, often fragmented outcomes toward locally ordered twisted states, while increasing communication connectivity drives them toward global synchronisation, and low spatial heterogeneity does not guarantee global consensus—40% of trials with Delta Z below 0.03 retain a twisted configuration through the final 20 turns.
University of Pennsylvania Shibaura Institute of Technology
Machiko Hirota（First author）
Multi-Agent LLM Systems LLM Agents Consensus Dynamics in Multi-Agent LLM Panels
arXiv Oct 2, 2026
HeteroFold lets Llama, Qwen, and Ministral share KV caches across families, running about 10.7x faster than native prefill at 32K context
The work proposes HeteroFold, a prefill-free cross-family KV cache transfer method that keeps both sender and receiver frozen and, through token alignment over shared character boundaries, cross-layer and cross-head K/V mapping, and receiver-aware calibration, achieves the best cache-transfer performance on all four long-context benchmarks across six transfer directions among Llama-3.1-8B, Qwen3-4B, and Ministral-3-14B, matches text-based communication on the HiddenBench multi-agent benchmark, and at 32K context makes Llama-3.1-8B to Ministral-3-14B transfer about 10.7x faster than native prefill.
The work proposes HeteroFold, a prefill-free cross-family KV cache transfer method that keeps both sender and receiver frozen and, through token alignment over shared character boundaries, cross-layer and cross-head K/V mapping, and receiver-aware calibration, achieves the best cache-transfer performance on all four long-context benchmarks across six transfer directions among Llama-3.1-8B, Qwen3-4B, and Ministral-3-14B, matches text-based communication on the HiddenBench multi-agent benchmark, and at 32K context makes Llama-3.1-8B to Ministral-3-14B transfer about 10.7x faster than native prefill.
New York University University of Southern California Seoul National University
Vincent-Daniel Yun（First author）
HeteroFold KV Cache Transfer KV Cache Reuse Multi-Agent LLM Systems
arXiv Oct 2, 2026
Labeling agents with their model family split multi-agent groups along the label, raising rounds and tokens and cutting success from 96% to 81%
Across two no-stakes cooperative games (Leader Election and Exclusion) and the GPQA-Diamond reasoning benchmark, with nine to twenty-five agents drawn from up to five open-weight model families, the study finds that when agents know each other's model family the interaction graph partitions into factions along that label, that the split follows the visible label rather than the underlying architecture (it persists under shuffled labels and neutral color tags and disappears when labels are removed), and that in strictly cooperative tasks labeled groups spend on average 30% more rounds and 55% more tokens to reach a decision while success drops from 96% to 81%.
Across two no-stakes cooperative games (Leader Election and Exclusion) and the GPQA-Diamond reasoning benchmark, with nine to twenty-five agents drawn from up to five open-weight model families, the study finds that when agents know each other's model family the interaction graph partitions into factions along that label, that the split follows the visible label rather than the underlying architecture (it persists under shuffled labels and neutral color tags and disappears when labels are removed), and that in strictly cooperative tasks labeled groups spend on average 30% more rounds and 55% more tokens to reach a decision while success drops from 96% to 81%.
Sapienza University of Rome
Xavier Del Giudice（First author）
Factionalism Multi-Agent LLM Systems Coordination games GPQA-Diamond

You’re all caught up

TGCC uses cross-layer trust gating so a compromised LLM agent self-revokes within steps instead of escalating

Across 50 stateless LLM agents updating predictions from locally visible peers, the study identifies synchronised, twisted, and chimera-like collective regimes, finding reasoning effort and communication topology govern different aspects of coordination

HeteroFold lets Llama, Qwen, and Ministral share KV caches across families, running about 10.7x faster than native prefill at 32K context

Labeling agents with their model family split multi-agent groups along the label, raising rounds and tokens and cutting success from 96% to 81%