paperarXivTrust 82 · PrimaryPublished 1mo agoLive · 1mo ago
A Lifecycle and Application-Stack Survey of Large Language Model Vulnerabilities: Attacks, Risks, Defenses, and Open Problems
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, prompts, model outputs, tools, memories, and user authority interact. This paper systematizes the literat
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- LinkedLinked via unknownIEEE Rolls Out Large Language Models Virtual Training Course →
- LinkedLinked via unknown"Dangerous" AI models are coming no matter what →
- LinkedLinked via unknownBook Review: Domain-Specific Small Language Models by Guglielmo Iozzia →
- PossiblePossibly related (embedding) · 47%aallan/vera →
- PossiblePossibly related (embedding) · 50%Pipelex/pipelex →
Covers
newsIEEE Rolls Out Large Language Models Virtual Training CoursenewsPrompt injection is exploiting enterprise AI's biggest design flaws by targeting agents, RAG pipelines and model routersnewsDefending against Prompt Injection with Structured Queries (StruQ) and Preference Optimization (SecAlign)news"Dangerous" AI models are coming no matter whatnewsBook Review: Domain-Specific Small Language Models by Guglielmo Iozzia
Covers (incoming)
newsA system-level approach to prompt injection: separating instruction and data channels in LLM agents [P]newsBest models for generating red-team attacks? Also looking for public datasets [R]newsManaging third-party model risk and AI dependencies - Security BoulevardnewsFreeBSoD: Leveraging Language Models to Find and Exploit Kernel Bugs (Part 2 of 2) - Security BoulevardnewsOver 250,000 Potential Security Issues Uncovered in GitHub Actions Workflows - TecheconomynewsLarge Language Models Misinformation: Ecosystem Security Shift - The CryptonomistnewsHow Large Language Models Decide Which Tools to Recommend — and What Developers Can Do About It - Programming InsidernewsIntroducing Antares: Highly Efficient Open Weight AI Models for Vulnerability Localization - Cisco BlogsnewsBSidesSF 2026 – Increasing The Analysis Surface Of Large Language Models - Security Boulevard
Implements (incoming)
Related across the graph
repolone-cloud/gerbilnewsBest models for generating red-team attacks? Also looking for public datasets [R]repozircote/rlm-rsnewsIntroducing Antares: Highly Efficient Open Weight AI Models for Vulnerability Localization - Cisco BlogsnewsPrompt injection is exploiting enterprise AI's biggest design flaws by targeting agents, RAG pipelines and model routersrepovicentereig/dspy.rbnewsHow Large Language Models Decide Which Tools to Recommend — and What Developers Can Do About It - Programming InsidernewsDefending against Prompt Injection with Structured Queries (StruQ) and Preference Optimization (SecAlign)newsOver 250,000 Potential Security Issues Uncovered in GitHub Actions Workflows - TecheconomyrepoPipelex/pipelexrepoGreyDGL/PentestGPTnewsBook Review: Domain-Specific Small Language Models by Guglielmo IozzianewsManaging third-party model risk and AI dependencies - Security BoulevardnewsLarge Language Models Misinformation: Ecosystem Security Shift - The Cryptonomistrepohuhusmang/Awesome-LLMs-for-Vulnerability-DetectionnewsBSidesSF 2026 – Increasing The Analysis Surface Of Large Language Models - Security BoulevardnewsFreeBSoD: Leveraging Language Models to Find and Exploit Kernel Bugs (Part 2 of 2) - Security BoulevardnewsIEEE Rolls Out Large Language Models Virtual Training CoursenewsA system-level approach to prompt injection: separating instruction and data channels in LLM agents [P]news"Dangerous" AI models are coming no matter whatrepovulnerability-lookup/VulnTrainrepoaallan/vera