Large Language Models (LLMs) are reshaping interaction on the Web, enabling users to express goals in natural language rather than navigating predefined application flows. However, conversational interfaces lack structure and affordances for direct manipulation. On the other hand, GUIs struggle to accommodate fluid, intention-driven interaction. This article discusses design trade-offs that can assist this Web transition toward agent-mediated, goal-driven paradigms and models, and proposes a unifying conceptual framework clarifying how agents can interpret user goals and assemble Web interfaces. This contribution outlines a shift toward goal-driven computing on the Web, with implications for design, interoperability, and agentic architectures.
The Internet of Things (IoT) is rapidly evolving toward increasingly intelligent, autonomous, and interconnected ecosystems, creating new challenges in security, reliability, and trustworthiness. This special issue brings together recent advances in securing next-generation IoT systems, with a focus on intelligent digital twin technologies and lightweight real-time anomaly detection frameworks. The selected articles demonstrate how artificial intelligence, edge computing, virtualization, and adaptive system design can be leveraged to improve the security and resilience of modern IoT environments. Collectively, these contributions highlight emerging directions for building scalable, robust, and deployment-orientated IoT security solutions and provide insights into the future evolution of secure and trustworthy IoT infrastructures.
Large Language Models (LLMs) are AI systems that learn statistical patterns in text to generate sentences, answer questions, and support conversations. Across an LLM life cycle, data from many individuals may be used during training or through user interactions. Since people have rights over their data, these rights should be respected. However, LLMs face practical and technical limits in enforcing them. This paper briefly explains how LLMs work, outlines key challenges for protecting individual rights, and reviews existing solutions.
Immersive computing is emerging as the next-generation computing paradigm, yet current systems are hindered by fragmented multimodal intelligence, siloed design, rigid infrastructure, and insufficient guarantees of trust and application grounding. This article presents a vision for artificial intelligence (AI)-native immersive computing, in which AI is embedded as a first-class primitive throughout the entire system stack. We propose a research agenda structured around three synergistic pillars to bridge these gaps. It prioritizes the creation of multimodal foundation models capable of unifying diverse data representations, performing adaptive inference in real time, and generating realistic 3-D world content. This intelligence is sustained by networked systems and infrastructure that leverage learning-driven orchestration and real-time coordination across scalable, shared platforms. To translate these advancements into meaningful real-world impact, the framework incorporates trustworthy, domain-specific application design by embedding end-to-end privacy, ensuring robust, reliable performance, and centering on task-centric operational requirements.
This article examines the governance challenges posed by autonomous artificial intelligence (AI) agents in democratic systems, arguing that they constitute a new form of institutionalization that reshapes how public decisions are made, legitimized, and contested. Drawing on institutional theory and algorithmic institutionalism, we show that AI agents do not merely automate tasks but introduce a distinct mode of behavior that can displace core democratic norms. The argument proceeds in three steps: first, we conceptualize AI agents as institutional actors embedded in goal-oriented systems; second, we show how their deployment creates a new, opaque layer of governance that competes with existing democratic structures; and third, we identify the key governance imperatives required to align these algorithmic institutions with accountability, participation, and legitimacy.
The growing ubiquity of cyberphysical systems (CPSs) embedded in the Internet of Things (IoT)–edge–cloud (IEC) continuum is transforming how data-driven applications are deployed and operated. Existing approaches to fault detection, diagnosis, and healing in such CPS deployments predominantly rely on centralized or statically supervised machine learning models. Due to the growing complexity of CPS systems, such as autonomous vehicles and smart cities, which require time-sensitive responses and utilize resource-constrained IoT and edge devices, fault detection presents several formidable research challenges. In this regard, multiagentic artificial intelligence, coupled with lifelong learning, offers a promising foundation. Despite its promise, realizing such decentralized and intelligent fault-management paradigms becomes not just beneficial, but necessary.
Entanglement-based quantum key distribution (QKD) offers robust key distribution through intrinsically correlated measurements, ensuring that any eavesdropping attempt introduces detectable disturbances. Although practical implementations are primarily limited to fixed fiber networks, extending QKD to mobile platforms like drones or small satellites introduces challenges in optical alignment, weight, and power consumption. This work outlines the core components and architecture for implementing BBM92-based QKD in mobile settings. A compact polarization Sagnac interferometer that utilizes a periodically poled potassium titanyl phosphate crystal provides a stable and efficient method for generating entangled photon pairs. The supporting subsystems include a polarization compensation system, detector module, and an acquisition, pointing, and tracking (APT) system. To assess performance under realistic conditions, we simulate photon loss, detector efficiency, and quantum bit error rates (QBERs) using the Simulator for Quantum Networks and Channels framework. The results show an average key generation rate of 5.43 bits/s, a final secure key length of 45.27 bits per block, and a QBER of 4.5%.
Quantum key distribution (QKD) offers provably secure, information-theoretic key exchange, but in long-distance scenarios without quantum repeaters, trusted nodes (TNs) are commonly employed despite introducing critical security risks. We propose a redundant key management method for QKD networks that combines twin-field QKD (or measurement-device independent QKD) with a novel key-routing scheme to eliminate the need for truly TNs. Quantum measurements are handled entirely within the network, minimizing end-user hardware requirements. Multiple QKD links connect intermediate nodes such that a successful attack requires the collusion of multiple adversarial nodes, greatly enhancing security over the traditional TN model. In this contribution, we discuss the tradeoff among security, key rates, and distances supported by the new method. Our analysis reveals that the improved redundant key management system may enable true end-to-end connectivity over several thousand kilometers while maintaining high security standards.
Reinforcement learning (RL) enables Internet of Things (IoT) systems to adapt to dynamic environments, but ensuring safety during learning is challenging. Fallback-based safe RL (SafeRL) methods guarantee safety by overriding unsafe actions but hinder learning in high-risk yet informative states and may disrupt previously learned safe behaviors. We propose RA-SafeRL, a risk-aware SafeRL framework that improves learning without compromising safety. RA-SafeRL uses a digital twin environment to safely explore high-risk actions and learn from them, applies adaptive fallback penalties reflecting action severity to distinguish risk levels, and reuses fallback trajectories via off-policy learning, converting previously discarded data into useful signals. Experiments show that RA-SafeRL outperforms baseline SafeRL by improving task performance by more than 30%, reducing fallback frequency by more than 10%, and shortening recovery time by 38%, enhancing safety, efficiency, and robustness in safety-critical IoT applications.
The intelligence deployed within RAN and CN largely operates independently, restricting the overall efficiency and adaptability of mobile networks, as key contextual data–such as real-time radio conditions and user session information–remains confined within a single domain. This article summarizes existing efforts from 3GPP, the O-RAN Alliance, and ITU-T, aimed at enabling intelligence integration, highlighting their current limitations. To address these limitations, we propose a RAN-CN intelligence convergence framework that bridges intelligence across the two domains. As a case study, we apply the proposed framework to conditional handover and show its potential to support target gNB selection. Finally, we discuss future research directions for intelligence-integrated mobile networks.
Microservices must meet end-to-end latency SLOs while optimizing per-service resources, but heterogeneous request paths make mapping resources to latency difficult. We propose DRvalve, a dual-layer framework that decouples applicationlevel SLO feedback from service-level control via performance targets, driven by a novel objective integrating CPU utilization and throttling rates. A global deep RL controller periodically sets CPU throttling targets, and local allocators scale resources using robust rollback and self-healing mechanisms to mitigate DRL lag and suppress cascading oscillations. Moreover, median denoising and binary-tree buffers ensure stable training and O(log n) efficiency. Evaluations on three production-trace workloads show that DRvalve saves up to 29.55% CPU over the best baseline and up to 52.67% over all baselines while strictly maintaining SLO compliance.
Generative AI tools have increased the scale and accessibility of deceptive image creation, but the robustness of existing forgery detection methods under emerging editing tools and real-world social media conditions remains uncertain. This study evaluates state-of-the-art image forgery detectors on recent benchmarks and applies the strongest model to a historical influence-campaign case study: Russian troll activity during the 2016 U.S. presidential election. We find that TruFor is the most consistent detector across legacy and recent datasets, but that its performance degrades on edits produced by advanced tools such as Nano Banana. We then deploy a multimodal pipeline combining forgery localization, Optical Character Recognition (OCR), face recognition, and visual-language-model-based annotation to characterize manipulated images in the campaign. Our analysis suggests that Russian troll accounts relied primarily on low-effort meme-style edits and ridicule-oriented reputation attacks, disproportionately targeting prominent U.S. political figures including Hillary Clinton, Bill Clinton, and Barack Obama.
This second special issue on quantum networking brings together four contributions that collectively capture the state of the art, spanning theory, system design, and experimental implementation.
Large language models excel at complex reasoning and natural language processing. However, probabilistic next-token prediction makes them inherently vulnerable to factual hallucinations. This unpredictability prevents their unmonitored use in highly regulated, mission-critical enterprise environments. To support high-assurance autonomous execution, we propose shifting from context engineering to knowledge graph (KG) engineering. We introduce the symbolic-in-the-loop (SITL) verification framework, a novel neurosymbolic architecture. Traditional retrieval augmented generation uses KGs for pregeneration context. Instead, SITL repositions the KG at the end of the pipeline as a formal verification layer. This postgeneration symbolic check reliably grounds probabilistic neural output into a deterministic ontology. The architecture seamlessly accommodates dynamic KG updates. This ensures validation against the most current symbolic state of an enterprise. We evaluate SITL across applications ranging from personal automation to multiagent orchestration. Experiments show this terminal KG approach successfully eliminates semantic drift. Ultimately, SITL ensures provably accurate outputs, replacing traditional human-in-the-loop models with scalable automation.
The widespread use of AI services has raised concerns for its environmental sustainability, towards which recent studies have identified carbon emissions of AI inference as the major contributor. This paper introduces a framework for designing AI inference incentives based on the users’ valuation for inference quality and latency, together with their environmental consciousness, while accounting for the tradeoff between carbon emissions and the two QoE parameters. Our approach can accommodate different tradeoffs, that depend on the size and complexity of the AI models and the allocation of resources to serve inference requests. The incentives can be offered through a practical two-tier service subscription that offers users a discount in exchange for reduced carbon emissions. The discounted service option gives the AI provider the flexibility to serve some percentage of inference requests at a lower quality and higher latency during periods of high carbon intensity.
Entanglement-based quantum key distribution (QKD) offers robust key distribution through intrinsically correlated measurements, ensuring any eavesdropping attempt introduces detectable disturbances. Although practical implementations are primarily limited to fixed fiber networks, extending QKD to mobile platforms like drones or small satellites introduces challenges in optical alignment, weight, and power consumption. This work outlines the core components and architecture for implementing BBM92-based QKD in mobile settings. A compact polarization Sagnac interferometer utilizing a PPKTP crystal provides a stable and efficient method for generating entangled photon pairs. Supporting subsystems include a polarization compensation system, a detector module, and an acquisition, pointing, and tracking (APT) system. To assess performance under realistic conditions, we simulate photon loss, detector efficiency, and quantum bit error rates using the SQUANCH framework. Results show an average key generation rate of 5.43 bits/s, a final secure key length of 45.27 bits per block, and a QBER of 4.5%.
Quantum Key Distribution (QKD) offers provably secure, information-theoretic key exchange, but in long-distance scenarios without quantum repeaters, Trusted Nodes (TNs) are commonly employed despite introducing critical security risks. We propose a redundant key management method for QKD network that combines Twin Field QKD (TF-QKD) (or Measurement-Device Independent (MDI)-QKD) with a novel key-routing scheme to eliminate the need for truly trusted TNs. Quantum measurements are handled entirely within the network, minimizing end-user hardware requirements. Multiple QKD links connect intermediate nodes such that a successful attack requires the collusion of multiple adversarial nodes, greatly enhancing security over the traditional TN model. In this contribution, we discuss the tradeoff between security, key rates, and distances supported by the new method. Our analysis reveals that the improved redundant key management system may enable true end-to-end connectivity over several thousand kilometers while maintaining high security standards.
Quantum networking exploits superposition and entanglement to enable secure communication, distributed computing, and advanced sensing. As the field moves toward practical realization, it requires interdisciplinary advances spanning physics, computer science, and network engineering, along with system-level innovations in architecture, protocols, and resource management. This special issue of IEEE Internet Computing highlights recent progress in the design, implementation, and application of quantum networks.
The computing continuum (CC) is expected to improve the quality of service of distributed applications. However, adding more devices to current cloud datacenters to implement the CC can increase the energy-related CO2 emissions of the paradigm. Sustainability requires intelligently managing applications through the CC, considering both the energy consumption and carbon intensity of energy sources. Moreover, the CC provider should perform this management, given their control of devices and networks. This work provides an overview on achieving sustainability in the CC and the associated challenges, proposing an architecture that leverages active inference to learn, manage, and reconfigure the system to achieve a low carbon footprint while meeting the service level objectives of applications.
The rise of quantum communication and networking heralds a transformative opportunity: distributed computing over quantum networks. This article offers a concise overview of the evolution of quantum communication and networking, highlights fundamental technologies and challenges, explores the efforts on real-world quantum network testbeds, and outlines our perspectives and ongoing efforts in this promising direction.
