EDITORIAL: Prompts: The Double-Edged Sword Using AI

Jordi Vallverdú^1,2*Rafal Rzepka³Alger Sans⁴

• ¹Autonomous University of Barcelona, Barcelona, Spain
• ²Institucio Catalana de Recerca i Estudis Avancats, Barcelona, Spain
• ³Hokkaido Daigaku, Sapporo, Japan
• ⁴Centre Nacional de Supercomputacio, Barcelona, Spain

broader conceptual landscape that motivated us to curate a collection addressing prompting from multiple disciplinary angles.Beyond these conceptual motivations, our collective expertise as Topic Editors also shaped the design of this Research Topic. Drawing from Rzepka's long-standing work in affective computing, computational linguistics, and human-machine interaction (e.g., Higuchi et al., 2008;Ptaszynski et al., 2009), and Sans Pinillos' research on abductive reasoning and the ethics of AI systems (e.g., Sans and Casacuberta, 2018), as well as its implications for dual-use technologies (e.g., Sans Pinillos and Vallverdú, 2025), we aimed to push the conversation one step further. Our intention was to move beyond the technical mechanics of prompting and to explore its broader epistemic, social, and creative implications. This interdisciplinary perspective allowed us to curate contributions that not only analyze prompting as it exists today but also envision how it may evolve in the near future, encouraging innovative and responsible uses of generative technologies.At a more global level, prompting itself is emerging as a new layer of computational technology. Recent work in natural language processing conceptualizes prompting as a new programming paradigm for large models, in which natural language becomes a high-level control language over pre-trained systems (e.g., Liu et al., 2022;White et al., 2023). From this perspective, prompts function less like ad hoc queries and more like an interface technology comparable to an operating system or an API. Treating prompting as such a foundational layer motivates the need for careful analysis of its epistemic, ethical, and creative dimensions, which is precisely the aim of the present Research Topic. The contribution GAAPO: genetic algorithmic applied to prompt optimization by Sécheresse, Guilbert-Ly, and Villedieu de Torcy illustrates a growing methodological trend: using computational tools to optimize prompts systematically. Their genetic algorithm demonstrates that prompt engineering can be automated, revealing formulations that significantly enhance performance. This raises important questions. As prompts become optimized by machines rather than humans, do we risk separating operational effectiveness from human interpretability? While automated discovery expands the expressive power of LLMs, it may also widen the gap between user understanding and model behavior. This tension is emblematic of the technical duality of prompting: it is both an accessible interface and a sophisticated control surface. Farnós, Sans Pinillos, and Costa, in Ethical prompting: toward strategies for rapid and inclusive assistance in dual-use AI systems, analyze prompting through the lens of ethics and governance. Prompts can enhance safety by enabling explicit constraint formulations, but they can also inadvertently bypass safeguards when poorly specified or intentionally manipulated. As models proliferate in sensitive or high-impact contexts, prompting becomes an ethical act, not merely an operational one. The authors argue compellingly for developing strategies that enable rapid and useful assistance while maintaining inclusivity and avoiding misuse. This aligns with broader discussions in AI governance: prompting increasingly resembles a form of literacy, where understandings of risk, bias, and responsibility must be integrated with technical competence, and where large language models are increasingly analyzed as sociotechnical systems whose scale and opacity raise concerns about bias, misuse, and environmental impact (e.g., Bender et al., 2021;Bommasani et al., 2021). Casacuberta and Guersenzvaig, in their article Disembodied creativity in generative AI: prima facie challenges and limitations of prompting in creative practice, examine prompting within artistic contexts.Generative systems enable new forms of creative production, yet the language-based nature of prompting introduces constraints. Much of creative practice relies on tacit, embodied, or material knowledgeelements difficult or impossible to encode as text. Their analysis shows that prompting can simultaneously open and limit creative spaces. Although generative models provide new expressive tools, they also risk standardizing artistic output around what is easily described. This reflects the deeper challenge of disembodied generative systems: they simulate meaning and creativity through linguistic coherence rather than experiential grounding. Across the contributions, three unifying themes emerge:1. Prompts as operational controls Prompts determine how systems behave, which capabilities are activated, and how models respond to uncertainty. Prompts shape what the model considers relevant, how it assembles explanations, and how it constructs apparent meaning. These dynamics resonate with earlier reflections on causal prompting and disembodied understanding.3. Prompts as socio-ethical instruments Because prompting can amplify or reduce risks, its role in dual-use scenarios must be carefully managed. Ethical prompting becomes indispensable in domains where trust, safety, and fairness matter. These themes clarify why prompting is inherently "double-edged." It democratizes AI access while introducing new vulnerabilities. It enhances creativity while imposing linguistic constraints. It provides powerful control over generative systems while making accountability more complex. Looking ahead, several research avenues appear especially urgent as prompting becomes further embedded in research, education, creativity, governance, and everyday technological practices. One key priority is the development of prompt literacy, ensuring that users not only learn how to obtain effective outputs but also understand the ethical, cultural, and epistemic dimensions embedded in each interaction with generative systems. Closely related to this is the need for explainable prompting: advancing methods that reveal why certain prompts succeed or fail, how optimized prompts differ from human-generated ones, and how users can maintain agency when interacting with increasingly opaque systems.Another important direction concerns multimodal and embodied prompting. Future AI systems may integrate textual instructions with perceptual, sensorimotor, or environmental cues, thereby reducing the overreliance on language alone and enabling richer forms of interaction. At the same time, increasing attention must be given to cultural and linguistic pluralism, as prompting practices vary significantly across languages and communities. Understanding these differences is vital not only for fairness and accessibility but also for preserving epistemic diversity in AI-mediated reasoning.Finally, prompting is poised to play a growing role in governmental and institutional decision-making.As public administrations explore the integration of AI-assisted tools into their workflows, prompting