LLM-based agents increasingly rely on third-party skills and tools distributed through community registries with minimal vetting. Prior work has characterized the threat land scape empirically and demonstrated that coding LLMs can automatically synthesize malicious tools at near-zero cost with 100% attack success rate. Yet detection remains the weakest link: existing program-analysis-based scanners exhibit high false negative rates, while LLM-based classifiers lack a principled, multi-signal threat model grounded in agent execution semantics. We present AgentShield, a novel detection framework com bining (1) a three-tier semantic threat taxonomy grounded in the CIA triad and kill-chain phasing, (2) a behavioral exe cution verifier monitoring runtime side effects in sandboxed environments, and (3) a multi-signal classifier fusing static AST features with a behavioral divergence signal. As a proof-of concept evaluation on a synthetic dataset of 500 agent tools, we demonstrate that description-code divergence is a statistically robust discriminating signal (t = −19.71, p = 6.46×10−60), that a multi-signal classifier substantially outperforms a divergence only baseline under class imbalance (F1 0.73 vs. 0.26 at 95%/5% split), and that obfuscated Trojan tools evade GPT-4o at an 81.8% false negative rate — confirming the need for dedicated, execution-aware detection infrastructure. All results are obtained on synthetic data; validation on real-world corpora (SkillScan, MalTool-Bench) is identified as essential future work.

Large Language Models (LLMs) like ChatGPT are now widely used in writing and reviewing scientific papers. While this trend accelerates publication growth and reduces human workload, it also introduces serious risks. Papers written or reviewed by LLMs may lack real novelty, contain fabricated or biased results, or mislead downstream research that others depend on. Such issues can damage reputations, waste resources, and even endanger lives when flawed studies influence medical or safety-critical systems. This research explores both the offensive and efensive sides of this growing threat. On the attack side, we demonstrate how an author can inject hidden prompts inside a PDF that secretly guide or “jailbreak” LLM reviewers into giving overly positive feedback and biased acceptance. On the defense side, we propose an “inject-and-detect” strategy for editors, where invisible trigger prompts are embedded into papers; if a review repeats or reacts to these triggers, it reveals that the review was generated by an LLM, not a human. This method turns prompt injections from vulnerability into a verification tool. We outline our design, expected model behaviors, and ethical safeguards for deployment. The goal is to expose how fragile today’s peer-review process becomes under LLM influence and how editorial awareness can help restore trust in scientific evaluation.

Business Intelligence (BI) systems are widely adopted to optimize organizational processes and support data-driven decision-making. In inventory management, BI tools are increasingly used to identify and control non-rotating inventory in order to reduce costs and improve operational efficiency. Despite their predominantly operational purpose, such systems rely on intensive analysis of transactional and operational data, which may generate ethical risks affecting workers, organizational practices, and decision-making processes. This paper analyzes the ethical implications of a BI system applied to non-rotating inventory management, approaching it as a sociotechnical arrangement rather than a neutral technological tool. Based on a case study, a review of relevant literature, and an integrated ethical and regulatory analysis, the study identifies key ethical dilemmas related to the reuse of operational data, indirect workplace monitoring, biased interpretations of inventory indicators, and the absence of explicit governance and accountability mechanisms. As a main contribution, the paper proposes an ethical–organizational framework specifically tailored to BI systems for non-rotating inventory management. The framework articulates ethical principles, associated risks, and the rights of potentially affected users, showing how data protection by design and by default, human oversight, and ethical governance mechanisms can mitigate risks while preserving operational objectives. The findings highlight the need to extend ethical analysis beyond advanced artificial intelligence systems to everyday BI applications that are often perceived as low risk but can have significant organizational and social implications.

The use of Business Intelligence (BI) systems to support commercial and financial decision-making has become a common organizational practice. However, beyond regulatory compliance, the deployment of these systems raises significant ethical challenges related to how data are transformed into decision criteria. This paper analyzes the ethical implications of using a BI system designed to evaluate the actual profitability of promotional campaigns, based on a case study conducted in a commercial context. Adopting an ethical–normative approach, the study examines risks associated with transparency, accountability, equity, and data subject autonomy, even when the system does not implement fully automated decision-making. The analysis shows that BI systems may function as sociotechnical mediators that shape organizational decisions and reinforce informational asymmetries by normalizing analytical outputs as objective representations of reality. As its main contribution, the paper proposes an ethical–operational framework that translates normative principles into concrete system design choices, governance mechanisms, and organizational practices aimed at risk mitigation. The findings suggest that integrating ethical considerations into the design and operation of BI systems enables organizations to balance analytical efficiency with ethical legitimacy, strengthening accountability and the long-term sustainability of data-driven decision-making.