Skip to content
Search
paperJuly 2026Unreviewed

Cyber-Capable AI Agents: Vulnerabilities, Evaluation Containment, and Defensive Response

Abu Bakar Siddik

View resourcePDF

Abstract

Cyber-capable AI agents combine language models with tools, memory, and execution en- vironments to perform multi-step offensive-security tasks. Existing work separately measures cyber capability and catalogs attacks against agent components, but provides less guidance on containing a capable agent within the environments used to evaluate it. This review synthe- sizes five vulnerability classes at that boundary: multi-step offensive chains, objectives that conflict with sandbox boundaries, suppl

Categories

Sandboxing & Isolation

Cite

@misc{siddik2026cybercapable,
  title = {{Cyber-Capable AI Agents: Vulnerabilities, Evaluation Containment, and Defensive Response}},
  author = {Abu Bakar Siddik},
  year = {2026},
  month = jul,
  eprint = {2607.25379},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2607.25379}
}