A Criminology of Machines

While the possibility of reaching human-like Artificial Intelligence (AI) remains controversial, the likelihood that the future will be characterized by a society with a growing presence of autonomous machines is high. In fact, autonomous AI agents are already deployed and active across several industries and digital environments. This trajectory points to a progressive hybridization of society marked by new forms of social interaction at both micro and macro levels. Alongside traditional human-human and human-machine interactions, machine-machine interactions are poised to become increasingly prevalent. Given these developments, I argue that criminology must begin to address the implications of this transition for crime and social control. Drawing on Actor–Network Theory and Woolgar’s decades-old call for a sociology of machines — frameworks that acquire renewed relevance with the rise of AI foundation models and generative agents — I contend that criminologists should move beyond conceiving AI solely as a tool. Instead, AI agents should be recognized as entities with agency, understood as a multi-layered construct encompassing computational, social, and legal dimensions. Building on insights from the literature on AI safety, I thus examine the risks and challenges associated with the rise of multi-agent AI systems, proposing a dual taxonomy to characterize the channels through which interactions among AI agents may generate deviant, unlawful, or criminal outcomes. I then advance and discuss four key questions that warrant theoretical and empirical attention: (1) Can we assume that machines will simply mimic humans? (2) Will crime theories developed for humans hence suffice to explain deviant or criminal behaviors emerging from interactions between autonomous AI agents? (3) What types of criminal behaviors will be affected first? (4) How might this unprecedented societal shift impact policing? These questions form the core of this article, underscoring the urgent need for criminologists to theoretically and empirically engage with the implications of multi-agent AI systems for the study of crime and play a more active role in debates on AI safety and governance.

There are two types of artificial general intelligence (AGI) safety solutions: global and local. Most previously suggested solutions are local: they explain how to align or “box” a specific AI (Artificial Intelligence), but do not explain how to prevent the creation of dangerous AI in other places. Global solutions are those that ensure any AI on Earth is not dangerous. The number of suggested global solutions is much smaller than the number of proposed local solutions. Global solutions can be divided into four groups: 1. No AI: AGI technology is banned or its use is otherwise prevented; 2. One AI: the first superintelligent AI is used to prevent the creation of any others; 3. Net of AIs as AI police: a balance is created between many AIs, so they evolve as a net and can prevent any rogue AI from taking over the world; 4. Humans inside AI: humans are augmented or part of AI. We explore many ideas, both old and new, regarding global solutions for AI safety. They include changing the number of AI teams, different forms of “AI Nanny” (non-self-improving global control AI system able to prevent creation of dangerous AIs), selling AI safety solutions, and sending messages to future AI. Not every local solution scales to a global solution or does it ethically and safely. The choice of the best local solution should include understanding of the ways in which it will be scaled up. Human-AI teams or a superintelligent AI Service as suggested by Drexler may be examples of such ethically scalable local solutions, but the final choice depends on some unknown variables such as the speed of AI progress.

Purpose Customer loyalty in the hospitality sector represents a critical determinant of a business’s success and competitive advantage. This paper aims to review the conceptual foundations of customer loyalty, its significance and the strategic mechanisms through which it can be cultivated and measured. Specifically, going beyond traditional strategies, this paper attempts to explain the concept of customer loyalty in the era of new technologies, especially AI agents, underscore its criticality and outline effective strategies for its enhancement and retention. Design/methodology/approach This paper synthesizes existing customer loyalty literature and proposes a framework based on insights from business, psychology and computer science to help companies and policymakers anticipate the impact of artificial intelligence (AI) agents on customer loyalty and guide future research directions in this emerging domain. Findings Building on prior literature and key developments in the last decade, this paper advocates for embedding retention-centric loyalty strategies while incorporating the newest technology. A proposed framework highlights the strategic alignment of AI capabilities, specifically AI agents, with loyalty objectives, emphasizing the critical role of data-driven personalization in sustaining competitive advantage and deepening customer relationships. Research limitations/implications The findings are primarily derived from secondary data sources and theoretical models, suggesting a need for empirical testing in diverse hospitality settings. Future research could explore the impact of AI and AI agents on loyalty across different cultures and market segments. Practical implications Hospitality firms may need to adapt loyalty strategies to account for AI-mediated decision-making. This includes enhancing algorithmic visibility, reconfiguring loyalty programs to engage both customers and their digital agents and understanding how AI shapes perceptions of value, convenience and brand preference. Firms must consider whether loyalty is being directed toward the brand, the agent or the ecosystem in which both operate. Social implications The integration of AI agents into loyalty ecosystems may have broader social consequences, including the erosion of consumer autonomy, increased algorithmic bias and new forms of digital exclusion. These transformations raise questions about the ethics of automated loyalty systems, the transparency of decision delegation and the future role of human connection in service interactions. Originality/value This paper fills a gap in existing research by examining the integration of AI with customer loyalty strategies within the hospitality industry. It offers a new perspective on how AI and AI agents can be aligned with traditional loyalty frameworks to enhance customer engagement and relationship management. The insights presented contribute to a deeper understanding of the practical implications of AI in shaping future loyalty programs and provide a foundation for further academic exploration and practical application in the field.

The proliferation of Artificial Intelligence (AI) agents, characterized by their autonomy and capacity for independent decision-making, presents both unprecedented opportunities and novel security challenges. This research paper provides a comprehensive analysis of the security landscape surrounding AI agents, examining the unique vulnerabilities stemming from their inherent characteristics and the emerging threat vectors targeting these autonomous systems. We delve into a categorized framework of potential attacks, ranging from data poisoning and adversarial manipulation to physical tampering and exploitation of autonomy. Furthermore, we critically evaluate existing and propose novel mitigation strategies, encompassing secure development practices, robustness training, explainable AI techniques for monitoring, and the crucial role of ethical and regulatory frameworks. This paper contributes to the growing body of knowledge on AI security, offering insights for researchers, developers, and policymakers navigating the complexities of securing the autonomous future. Keywords: Artificial Intelligence, AI Agents, Autonomous Systems, Cybersecurity, Machine Learning Security, Adversarial Attacks, Data Poisoning, Robotics Security, Ethical AI, AI Governance

In the last years, artificial intelligence (AI) safety gained international recognition in the light of heterogeneous safety-critical and ethical issues that risk overshadowing the broad beneficial impacts of AI. In this context, the implementation of AI observatory endeavors represents one key research direction. This paper motivates the need for an inherently transdisciplinary AI observatory approach integrating diverse retrospective and counterfactual views. We delineate aims and limitations while providing hands-on-advice utilizing concrete practical examples. Distinguishing between unintentionally and intentionally triggered AI risks with diverse socio-psycho-technological impacts, we exemplify a retrospective descriptive analysis followed by a retrospective counterfactual risk analysis. Building on these AI observatory tools, we present near-term transdisciplinary guidelines for AI safety. As further contribution, we discuss differentiated and tailored long-term directions through the lens of two disparate modern AI safety paradigms. For simplicity, we refer to these two different paradigms with the terms artificial stupidity (AS) and eternal creativity (EC) respectively. While both AS and EC acknowledge the need for a hybrid cognitive-affective approach to AI safety and overlap with regard to many short-term considerations, they differ fundamentally in the nature of multiple envisaged long-term solution patterns. By compiling relevant underlying contradistinctions, we aim to provide future-oriented incentives for constructive dialectics in practical and theoretical AI safety research.

The rapid integration of artificial intelligence (AI) across economic, governmental, and societal systems has transformed decision-making, productivity, and innovation at unprecedented scale.As AI systems increasingly influence critical domains such as healthcare, finance, national security, and public administration, concerns regarding accountability, transparency, reliability, and harm mitigation have become central to global policy and technical discourse.From a broad perspective, effective AI adoption now depends not only on performance gains but also on the establishment of robust governance structures that align technological progress with ethical norms, legal frameworks, and societal values.Within this evolving landscape, AI governance provides the institutional and regulatory foundation for responsible development and deployment, defining roles, oversight mechanisms, and compliance obligations across the AI lifecycle.Building on governance, AI safety focuses on ensuring that systems behave as intended, remain aligned with human objectives, and minimize risks arising from model errors, bias, emergent behaviors, or misuse.Complementing safety, AI security controls address adversarial threats, data integrity, model robustness, and resilience against attacks such as data poisoning, model inversion, and prompt exploitation.This abstract narrow the discussion to the intersection of AI governance, AI safety, and AI security controls, emphasizing their interdependence as a unified risk management framework.Together, these pillars are essential for sustaining public trust, enabling innovation, and ensuring that AI systems remain secure, controllable, and beneficial at scale.

Human-AI collaboration: Unraveling the effects of user proficiency and AI agent capability in intelligent decision support systems

As artificial intelligence (AI) agents become deeply integrated into fitness systems, trustworthy human–AI agent interaction has become pivotal for user engagement in smart home fitness (SHF) e-commerce platforms. Grounded in the Computers Are Social Actors (CASA) framework, this study empirically investigates how, acting as AI fitness coaches, AI agents’ technical and social features shape users’ active engagement in the in-home social e-commerce context. A mixed-method approach was employed, combining computational text mining of 17,582 user reviews from fitness e-commerce platforms with a survey (N = 599) of Chinese consumers. The results show that (1) the technical–social features of AI agents serving as AI fitness coaches include visibility, gamification, interactivity, humanness, and sociability; (2) these five technical–social features of AI agents positively influence user compliance via both cognitive and emotional trust in AI agents; (3) these five technical–social features of AI agents serving as AI fitness coaches positively impact active engagement via both cognitive and emotional trust in AI agents. This study extends the CASA framework to the domain of AI coaching by demonstrating the parallel roles of cognitive and emotional trust in AI agents. For designers and managers in the fitness e-commerce industries, this study offers actionable insights for designing AI agents integrating functional and social features that foster trust and drive behavioral outcomes.

In the context of the rapid digitalization of higher education, proactive artificial intelligence (AI) agents embedded within multi-agent systems (MAS) offer new opportunities for personalized learning, improved quality of education, and alignment with sustainable development goals. This study aims to analyze how such AI solutions are perceived by students at Narxoz University (Kazakhstan) prior to their practical implementation. The research focuses on four key aspects: the level of student trust in AI agents, perceived educational value, concerns related to privacy and autonomy, and individual readiness to use MAS tools. The article also explores how these solutions align with the Sustainable Development Goals—specifically SDG 4 (“Quality Education”) and SDG 8 (“Decent Work and Economic Growth”)—through the development of digital competencies and more equitable access to education. Methodologically, the study combines a bibliometric literature analysis, a theoretical review of pedagogical and technological MAS concepts, and a quantitative survey (n = 150) of students. The results reveal a high level of student interest in AI agents and a general readiness to use them, although this is tempered by moderate trust and significant ethical concerns. The findings suggest that the successful integration of AI into educational environments requires a strategic approach from university leadership, including change management, trust-building, and staff development. Thus, MAS technologies are viewed not only as technical innovations but also as managerial advancements that contribute to the creation of a sustainable, human-centered digital pedagogy.

PurposeThis paper identifies the effects of different dimensions of the cuteness (i.e. baby schema cuteness and whimsical cuteness) of artificial intelligence (AI) agents on attachment to them. In addition, the current paper examines the consequences of the attachment to AI agents.Design/methodology/approachA pretest to validate the measurement scale for the attachment to AI agents and a survey study were conducted with AI agent users. The authors used structural equation modeling to analyze the data for hypothesis testing.FindingsThe baby schema and whimsical cuteness of AI agents drive consumers to develop stronger attachments to their AI agents. This is because consumers perceive cute AI agents as being more trustworthy. As a result, consumers who feel attached to their AI agents are more inclined to report higher satisfaction and commitment levels. They are also more likely to purchase products or services recommended by their AI agents and use them more frequently.Originality/valueDespite the growing popularity of AI agents, there is a lack of understanding regarding which characteristics of AI agents affect consumer behavior. Therefore, this research examines how the attribute of cuteness influences consumers' attachment to AI agents and subsequently affects their satisfaction and purchase intention toward products recommended by AI agents. Our study demonstrates that the element of cuteness in AI agents plays a crucial role in shaping perceptions of benevolence trustworthiness, as well as fostering users' attachment to AI agents. Furthermore, we observe positive consumer behaviors as a result of their attachment to AI agents. The findings from this study provide valuable insights for practitioners on how to effectively utilize cuteness in AI agents.

Background Large language models (LLMs) have gained attention for their ability to exhibit human-like clinical reasoning with mock clinical cases. However, because of privacy concerns, few studies have evaluated their use in real-world healthcare settings. We aimed to assess the accuracy of LLMs in auditing blood culture appropriateness using real charts.Prompt Provided to Initial Reviewer AI Agent for Blood Culture Appropriateness ClassificationAI agents were guided by structured inclusion and exclusion criteria to assess blood culture appropriateness. Prompts included clinical definitions, required supporting evidence, and explicit instructions to avoid assumptions or external reasoning beyond the documentation in the clinical note. Agents were also asked to provide quoted justification for their classifications. The criteria were adapted from the Johns Hopkins Prevention Epicenter Blood Culture Stewardship Collaborative algorithm and based on: Fabre V, Sharara SL, Salinas AB, Carroll KC, Desai S, Cosgrove SE. Does This Patient Need Blood Cultures? A Scoping Review of Indications for Blood Cultures in Adult Nonneutropenic Inpatients. Clin Infect Dis. 2020; PMID: 31942949.Prompt Provided to Double Checker AI Agent for Verification of Blood Culture Appropriateness AssessmentThe second AI agent was tasked with verifying the initial reviewer’s justification against explicit inclusion criteria for blood culture ordering. The prompt instructed the agent to identify whether the justification explicitly met at least one inclusion criterion and to flag assumptions or unsupported reasoning. Final classifications required quoting or rejecting specific evidence from the patient’s chart. Methods Stanford University deployed secure LLMs with direct access to electronic medical records. Using these, we developed two artificial intelligence (AI) agents—task-specific models designed to audit blood culture order appropriateness based on previously published criteria. We applied the agents to a random sample of 105 blood culture orders previously audited by an infectious diseases provider between May and December 2024. After excluding repeat orders within 48 hours, 67 unique cases remained (31 appropriate, 36 not). Each case included all assessment and plan notes from admission to blood culture collection (range: 1–500 notes). The initial reviewer agent (gpt-4o-mini; OpenAI) scanned the notes for any mention of appropriateness or non-appropriateness criteria. A second, more powerful double-checker agent (o1-mini; OpenAI) then reviewed and, if necessary, corrected the initial classification.Performance Metrics of Secure AI Agent Classifying Blood Culture Appropriateness — Northern California, 2025The AI agent's performance is summarized across six standard classification metrics. While sensitivity and negative predictive value were relatively high, specificity and precision were lower, reflecting a tendency to overflag orders as appropriate. The 0.5 reference threshold is marked for interpretability.Case-Level Examples of Blood Culture Orders Reviewed by AI Agents and Adjudicated for AppropriatenessEach row represents a single blood culture order with corresponding adjudication status, AI agent classifications, and rationale. Green indicates an appropriate order; red indicates an inappropriate order. AI explanations were based on clinical notes available prior to blood culture collection. Discrepancies between human adjudication and AI classification are highlighted in the initial and final classification columns. Results Overall performance of the AI agents was modest, with a balanced accuracy of 0.568, sensitivity of 0.774, and specificity of 0.361. The agents frequently over-flagged blood culture orders as appropriate, demonstrating a tendency to recommend blood cultures in a broad range of cases. This likely reflects a known LLM behavior, sycophancy, where the model aligns with the reasoning presented in the clinical notes, such as agreeing with the care team’s suspicion of sepsis, even when objective criteria were not met. Notably, the “severe sepsis/septic shock” criterion was the most common justification given by the AI agents for classifying orders as appropriate. Conclusion The AI agents demonstrated limited performance in adjudicating blood culture appropriateness. Their decisions were largely influenced by sycophantic bias and the presence of the word sepsis in the notes. Their utility in medical classification tasks may be best suited for initial screening rather than clinical recommendations. Disclosures All Authors: No reported disclosures

PurposeThe purpose of this paper is to explain to readers how intelligent systems can fail and how artificial intelligence (AI) safety is different from cybersecurity. The goal of cybersecurity is to reduce the number of successful attacks on the system; the goal of AI Safety is to make sure zero attacks succeed in bypassing the safety mechanisms. Unfortunately, such a level of performance is unachievable. Every security system will eventually fail; there is no such thing as a 100 per cent secure system.Design/methodology/approachAI Safety can be improved based on ideas developed by cybersecurity experts. For narrow AI Safety, failures are at the same, moderate level of criticality as in cybersecurity; however, for general AI, failures have a fundamentally different impact. A single failure of a superintelligent system may cause a catastrophic event without a chance for recovery.FindingsIn this paper, the authors present and analyze reported failures of artificially intelligent systems and extrapolate our analysis to future AIs. The authors suggest that both the frequency and the seriousness of future AI failures will steadily increase.Originality/valueThis is a first attempt to assemble a public data set of AI failures and is extremely valuable to AI Safety researchers.

Artificial intelligence (AI) agents are widely used in the retail and distribution industry. The primary objective was to investigate whether the gender of AI agents influences trust and grounding. This paper examined the influence of AI agent gender and brand concepts on trust and grounding within virtual brand spaces. For this purpose, it used two independent variables: brand concept (functional vs. experiential) and AI agent gender (male vs. female). The dependent variables included AI agent trust and grounding. The study revealed that in virtual brand spaces centered around a functional concept, male AI agents generated higher levels of trust than female AI agents, whereas, when focused on an experiential concept, female AI agents induced higher levels of grounding than male AI agents. Furthermore, the findings indicate that the association between customers’ identification with AI agents and recommendations for actual brand purchases is mediated by trust and grounding. These findings support the idea that users who strongly identify with AI agents are more inclined to recommend brand products. By presenting alternatives that foster the establishment and sustenance of a meaningful, sustainable relationship between humans and AI, this study contributes to research on human–computer interactions.

To develop an artificial intelligence (AI) agent for fully automated rapid head-and-neck intensity-modulated radiation therapy (IMRT) plan generation without time-consuming dose-volume-based inverse planning. This AI agent was trained via implementing a conditional generative adversarial network (cGAN) architecture. The generator, PyraNet, is a novel deep learning network that implements 28 classic ResNet blocks in pyramid-like concatenations. The discriminator is a customized four-layer DenseNet. The AI agent first generates multiple customized two-dimensional projections at nine template beam angles from a patient's three-dimensional computed tomography (CT) volume and structures. These projections are then stacked as four-dimensional inputs of PyraNet, from which nine radiation fluence maps of the corresponding template beam angles are generated simultaneously. Finally, the predicted fluence maps are automatically postprocessed by Gaussian deconvolution operations and imported into a commercial treatment planning system (TPS) for plan integrity check and visualization. The AI agent was built and tested upon 231 oropharyngeal IMRT plans from a TPS plan library. 200/16/15 plans were assigned for training/validation/testing, respectively. Only the primary plans in the sequential boost regime were studied. All plans were normalized to 44Gy prescription (2Gy/fx). A customized Harr wavelet loss was adopted for fluence map comparison during the training of the PyraNet. For test cases, isodose distributions in AI plans and TPS plans were qualitatively evaluated for overall dose distributions. Key dosimetric metrics were compared by Wilcoxon signed-rank tests with a significance level of 0.05. All 15 AI plans were successfully generated. Isodose gradients outside of PTV in AI plans were comparable to those of the TPS plans. After PTV coverage normalization, Dmean of left parotid (DAI =23.1±2.4Gy; DTPS =23.1±2.0Gy), right parotid (DAI =23.8±3.0Gy; DTPS =23.9±2.3Gy), and oral cavity (DAI =24.7±6.0Gy; DTPS =23.9±4.3Gy) in the AI plans and the TPS plans were comparable without statistical significance. AI plans achieved comparable results for maximum dose at 0.01cc of brainstem (DAI =15.0±2.1Gy; DTPS =15.5±2.7Gy) and cord+5mm (DAI =27.5±2.3Gy; DTPS =25.8±1.9Gy) without clinically relevant differences, but body Dmax results (DAI =121.1±3.9Gy; DTPS =109.0±0.9Gy) were higher than the TPS plan results. The AI agent needed ~3s for predicting fluence maps of an IMRT plan. With rapid and fully automated execution, the developed AI agent can generate complex head-and-neck IMRT plans with acceptable dosimetry quality. This approach holds great potential for clinical applications in preplanning decision-making and real-time planning.

Artificial Intelligence (AI) agents are revolutionizing industries by enabling autonomous decision-making, task execution, multi-agent collaboration. This paper provides a comprehensive review of AI agent frameworks, focusing on their architectures, applications, challenges in financial services. We conduct a comparative analysis of leading frameworks, including LangGraph, CrewAI, AutoGen, evaluating their strengths, limitations, suitability for complex financial tasks such as trading, risk assessment, investment analysis. The integration of AI agents in financial markets presents both opportunities challenges, particularly in terms of regulatory compliance, ethical considerations, model robustness. We examine agentic AI design patterns, multi-agent systems, the deployment of AI agents advancing the proposal to use them for fraud detection risk management. By synthesizing insights from academic research industry practices, this review identifies key trends future directions in AI agent development. This work contributes to the growing discourse on AI-driven automation by outlining technical considerations open challenges in deploying AI agents at scale. We highlight the need for enhanced transparency, interpretability, security in AI-driven Agentic systems. Our findings provide valuable insights for researchers practitioners seeking to harness AI agents for more efficient intelligent decision-making.