paperJune 2026Unreviewed

A Lifecycle and Application-Stack Survey of Large Language Model Vulnerabilities: Attacks, Risks, Defenses, and Open Problems

Seyed Bagher Hashemi Natanzi, Bo Tang

Abstract

Large language models are no longer only text generators. They are increasingly embedded in retrieval pipelines, enterprise assistants, coding environments, robotic systems, security-operation workflows, and autonomous agents that can read private data, call tools, write files, execute code, and act across organizational boundaries. This shift changes the security problem: risks do not arise from the model weights alone, but from the full lifecycle and application stack through which data, promp

Cite

@misc{natanzi2026lifecycle,
  title = {{A Lifecycle and Application-Stack Survey of Large Language Model Vulnerabilities: Attacks, Risks, Defenses, and Open Problems}},
  author = {Seyed Bagher Hashemi Natanzi and Bo Tang},
  year = {2026},
  month = jun,
  eprint = {2606.31639},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2606.31639}
}