Towards trustworthy agentic AI: a comprehensive survey of safety, robustness, privacy, and system security
arXiv:2605.23989v1 Announce Type: cross Abstract: Agentic AI systems -- Large Language Models (LLMs) augmented with planning, tool use, memory, and long-horizon interactions -- can execute complex tasks autonomously, but their multi-step trajectories introduce new failure modes that challenge trustworthiness. This survey provides a focused examination of trustworthy agentic AI through two core dimensions that are critical for high-risk deployments: Safety and Robustness, and Privacy and System Security. For each dimension, we clarify key concepts, identify where risks emerge along the agent wo
The rapid advancement and deployment of agentic AI systems necessitate a comprehensive understanding of their trustworthiness, especially as these systems transition to high-risk applications demanding robust safety and security measures.
This survey provides a critical framework for understanding the emerging risks and necessary safeguards for agentic AI, which is poised to automate complex tasks and workflows across various industries.
The focus on 'trustworthiness' introduces a new dimension to AI development, moving beyond pure capability to encompass safety, robustness, privacy, and system security as core requirements for widespread adoption and regulatory acceptance.
- · AI security firms
- · AI ethics researchers
- · Autonomous system developers
- · Cybersecurity sector
- · Unregulated AI developers
- · Companies with poor security practices
- · Traditional white-collar service providers
Increased investment in agentic AI safety and security research and development.
New regulatory frameworks and compliance standards will emerge for autonomous AI systems operating in critical sectors.
The development of 'trustworthy AI' certifications could become a strategic advantage, shaping market leadership and adoption rates.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL