LATIDIA · Investigación

Trustworthy Agentic AI: A Comprehensive Cybersecurity and Systems Survey on Threat Landscapes, Defense Architectures, and Open Challenges

arXiv: 2609.13731v1Tipo de anuncio: nuevo Resumen: La transición de los modelos de cimientos pasivos a los sistemas de IA agéntica autónomos y dirigidos a objetivos ha introducido capacidades sin precedentes al acoplar reas cognitivas recursivas

La noticia

arXiv:2609.13731v1 Announce Type: new Abstract: The transition from passive foundation models to autonomous, goal-directed agentic AI systems has introduced unprecedented capabilities by coupling recursive cognitive reasoning loops, persistent memory architectures, live tool execution planes, and multi-agent collaboration topologies. However, granting probabilistic neural cores execution authority across filesystems, networks, and cloud infrastructure dissolves classical security perimeters: natural language simultaneously serves as input data, internal control code, and communication protocols, exposing a Turing-complete blast radius where untrusted data represents executable instructions. This survey delivers a comprehensive systems-security reference framework for trustworthy agentic AI, synthesizing 206 foundational studies and regulatory standards. We formalize the general agent architecture as a stateful 5-tuple and establish