I. de Zarzà

Advanced Integration of Large Language Models for Autonomous Systems and Critical Decision SupportLarge language models (LLMs) have shown transformative potential in autonomous systems and critical decision-making, yet standalone models remain limited in robustness, reliability, and safety assurance when deployed in high-stakes environments. This Research Topic began from the premise that such limitations are better addressed by structured integration of multiple specialized models than by scaling any single one (Guo et al., 2024). We invited work on multi-LLM integration for perception, navigation, and decision-making in robots, drones, and vehicles; on human-robot collaboration; on high-stakes decision support; on verification, uncertainty quantification, and safety assurance; and on real-time adaptation.Seven contributions were accepted, spanning agent generation, orchestration, perception, query translation, automated machine learning, intrusion detection, and governance.Deployment in these settings changes the central question. Performance can no longer be judged by fluency or task accuracy alone; it must also be assessed through grounding, reproducibility, latency, calibration, failure containment, human oversight, and auditability. Across domains, the contributions converge on a common conclusion: dependable autonomy is principally a systems-engineering problem.Reliability emerges not from trusting a single model, but from structuring how models are composed, constrained, checked, and connected to action.Perera et al. challenge the fixed-team assumption of many multi-agent systems. Their Initial Automatic and Dynamic Real-Time Agent Generation mechanisms create specialized agents from evolving conversational context. In the evaluated medical scenario, dynamic generation improved coverage, lexical diversity, and thematic relevance over a static AutoGen configuration, treating system composition itself as an adaptive variable. The same move from fixed programs toward prompt-defined behavior appears in LLM-driven swarm simulations (Jimenez-Romero et al., 2025).1 de Zarz à et al.Zhou and Chan address the complementary problem of reproducibility. Their orchestrator, ORCH, 27 decomposes a problem, gathers analyses from heterogeneous models, and merges them through a 28 deterministic protocol; an optional exponential moving average module adapts routing from historical 29 feedback. Gains are strongest on harder reasoning tasks, but entail substantial latency and cost. Taken 30 together, these studies show that adaptation and determinism are not opposites: agent membership and 31 routing may while interfaces, rules, and aggregation procedures remain explicit and 32 auditable, as in ensemble-and-arbiter designs where inter-model disagreement is measured and routed to 33 human review (Lipianina-Honcharenko et al., 2026). Calboreanu makes the architectural argument most explicit. LATTICE separates planning, execution, and 58 governance so that no component both decides an action and judges compliance; it applies policy-as-code 59 through gated execution, escalates uncertain cases to human operators, and preserves provenance through The next phase should connect these principles into end-to-end assurance cases. It requires interoperable 78 agent-tool contracts, benchmarks covering distribution shift and adversarial faults (Radanliev et al., 2026), 79 selective autonomy with tested fallback behavior, and human-centered studies of explanation and escalation.The central lesson is measured but consequential: LLMs become suitable for autonomous systems and 81 critical decision support not as self-sufficient decision makers, but when embedded within architectures that 82 make uncertainty visible, constrain action, preserve accountability, and retain meaningful human control 83 (Santoni de Sio and van den Hoven, 2018).We thank all contributing authors, reviewers, and the Frontiers editorial team for advancing this 85 interdisciplinary discussion.

Introduction: Large language models (LLMs) can generate persuasive narratives that shift agent behavior in multi-agent systems, but deploying raw, unstructured text as an influence mechanism offers no formal guarantees on effectiveness, interpretability, or controllability. Methods: We introduce the LLM Influence Compiler (LIC), a solver–critic pipeline that compiles natural-language cooperation directives into structured, parameterized influence policies defined over a five-field schema: network targeting, narrative theme, intensity, deployment timing, and compiler confidence. Each compiled policy is diffused through a network of numerical agents via an exposure model incorporating fatigue decay, susceptibility heterogeneity, and backlash. Evaluation spans nine controlled experiment blocks comprising more than 200 simulation runs across four topologies (Barabási–Albert, small-world, Erdős–Rényi, modular SBM), four network sizes ( n ∈ {80, 160, 320, 640}), five LLM backbones, and two non-stationary perturbations. Results: Compiled policies raise the mean cooperation rate to 0.826 ± 0.010 (95% bootstrap CI [0.819, 0.834]), an 8.6% relative improvement over the unstructured baseline (0.760 ± 0.005, Mann–Whitney p _Bonf = 0.040, Cohen's d = 7.90). A controlled decomposition attributes the gain to network-aware targeting ( ca . +4.5 pp), dose calibration (+1.9 pp), and an intervention floor (+1.5 pp); the residual contribution of full LLM compilation over a hand-coded rule with the same parameters is statistically indistinguishable from zero under stationary conditions. Under structural non-stationarity (mid-simulation graph rewire), the LLM-mediated re-deployment architecture significantly outperforms a frozen rule (paired t-test p < 0.001, paired Cohen's d = 2.27, 8/8 seeds). Adversarial stress testing confirms that the critic correctly flags 20/20 risky policies as high risk and rejects them, while passing a moderate-baseline policy at medium risk in all 5/5 trials. Discussion: The compiler abstraction converts an opaque generative process into a decomposable, auditable policy object whose components can be independently attributed, compared, and governed, laying the groundwork for constitutional oversight of LLM-mediated influence in artificial societies.

Intrusion detection in industrial cyber-physical systems is constrained by small labelled-attack corpora and by the subtler signal of physical-process attacks compared with classical IT-network intrusions, motivating renewed interest in foundation-model-based detectors; classical detectors are typically trained per dataset and degrade under the distribution shift that is common in operational technology, where attack repertoires evolve faster than retraining cycles. Two foundation-model families are now plausible candidates: open-source Large Language Models (LLMs) and recent tabular foundation models (TabPFN, TabICL) pre-trained for in-context tabular inference. We compare the two families head-to-head, alongside Random Forest and XGBoost classical anchors, across three established industrial security benchmarks (SWaT, HAI, WUSTL-IIoT-2021) under a controlled multi-seed full-holdout protocol with paired McNemar and cross-seed Mann–Whitney tests. The empirical picture is dataset-dependent rather than universal: tabular foundation models establish a strong, previously unreported baseline that is competitive with or superior to classical anchors on every dataset evaluated, while LLMs are complementary detectors with a specific advantage on schemas that carry process-engineering semantics (such as SWaT’s named sensor channels). A per-class analysis on the WUSTL five-class attack taxonomy shows that the two families have structurally different strengths: tabular methods dominate traffic-rich attacks (Denial-of-Service, Reconnaissance), whereas LLMs are competitive on rare attack types (Backdoor, Command Injection). A confidence-gated cascade that escalates only low-confidence tabular decisions to an LLM exceeds either detector alone at a small query budget, and a leave-one-attack-type-out analysis shows that foundation-model detectors generalise to unseen attack families substantially better than the classical anchors. The appropriate detector choice in industrial cyber-physical security is therefore informed by the dataset’s feature schema, the attack-type mix, and the operational cost envelope, rather than by a specific performance metric.

This paper presents a comprehensive cross-platform evaluation of reasoning capabilities in contemporary foundation models, establishing an infrastructure-agnostic benchmark across three computational paradigms: HPC supercomputing (MareNostrum 5), cloud platforms (Nebius AI Studio), and university clusters (a node with eight H200 GPUs). We evaluate 15 foundation models across 79 problems spanning eight academic domains (Physics, Mathematics, Chemistry, Economics, Biology, Statistics, Calculus, and Optimization) through three experimental phases: (1) Baseline establishment: Six models (Mixtral-8x7B, Phi-3, LLaMA 3.1-8B, Gemma-2-9b, Mistral-7B, OLMo-7B) evaluated on 19 problems using MareNostrum 5, establishing methodology and reference performance; (2) Infrastructure validation: The 19-problem benchmark repeated on university cluster (seven models including Falcon-Mamba state-space architecture) and Nebius AI Studio (nine state-of-the-art models: Hermes-4 70B/405B, LLaMA 3.1-405B/3.3-70B, Qwen3 30B/235B, DeepSeek-R1, GPT-OSS 20B/120B) to confirm infrastructure-agnostic reproducibility; (3) Extended evaluation: Full 79-problem assessment on both university cluster and Nebius platforms, probing generalization at scale across architectural diversity. The findings challenge conventional scaling assumptions, establish training data quality as more critical than model size, and provide actionable guidelines for model selection across educational, production, and research contexts. The tri-infrastructure methodology and 79-problem benchmark enable longitudinal tracking of reasoning capabilities as foundation models evolve.

While traditional ensemble methods have dominated tabular intrusion detection systems (IDSs), recent advances in foundation models present new opportunities for enhanced cybersecurity applications. This paper presents a comprehensive multi-modal evaluation of foundation models—specifically TabPFN (Tabular Prior-Data Fitted Network), TabICL (Tabular In-Context Learning), and large language models—against traditional machine learning approaches across three cybersecurity datasets: CIC-IDS2017, N-BaIoT, and CIC-UNSW. Our rigorous experimental framework addresses critical methodological challenges through model-appropriate evaluation protocols and comprehensive assessment across multiple data variants. Results demonstrate that foundation models achieve superior and more consistent performance compared with traditional approaches, with TabPFN and TabICL establishing new state-of-the-art results across all datasets. Most significantly, these models uniquely achieve non-zero recall across all classes, including rare threats like Heartbleed and Infiltration, while traditional ensemble methods—despite achieving >99% overall accuracy—completely fail on several minority classes. TabICL demonstrates particularly strong performance on CIC-IDS2017 (99.59% accuracy), while TabPFN maintains consistent performance across all datasets, suggesting robust generalization capabilities. Both foundation models achieve these results using only fractions of the available training data and requiring no hyperparameter tuning, representing a paradigm shift toward training-light, hyperparameter-free adaptive IDS architectures, where TabPFN requires no task-specific fitting and TabICL leverages efficient in-context adaptation without retraining. Cross-dataset validation reveals that foundation models maintain performance advantages across diverse threat landscapes, while traditional methods exhibit significant dataset-specific variations. These findings challenge the cybersecurity community’s reliance on tree-based ensembles and demonstrate that foundation models offer superior capabilities for next-generation intrusion detection systems in IoT environments.

Traffic congestion and carbon emissions remain pressing challenges in urban mobility. This study explores the integration of UAV (drone)-based monitoring systems and IoT sensors, modeled as induction loops, with Large Language Models (LLMs) to optimize traffic flow. Using the SUMO simulator, we conducted experiments in three urban scenarios: Pacific Beach and Coronado in San Diego, and Argüelles in Madrid. A Gemini-2.0-Flash experimental LLM was interfaced with the simulation to dynamically adjust vehicle speeds based on real-time traffic conditions. Comparative results indicate that the AI-assisted approach significantly reduces congestion and CO₂ emissions compared to a baseline simulation without AI intervention. This research highlights the potential of UAV-enhanced IoT frameworks for adaptive, scalable traffic management, aligning with the future of drone-assisted urban mobility solutions.