SecPapers

A living, searchable catalog of large language model security research.

Explore the web index | Browse all papers | Use the dataset | Methodology | Suggest a paper

SecPapers tracks both sides of LLM security: research that makes language models safer, and research that applies language models to cybersecurity. It queries arXiv every day, applies a transparent relevance filter, deduplicates paper revisions, and regenerates this repository from stable source data.

At a glance

1148 papers across 4 publication years. Latest arXiv metadata update: 2026-09-08.

Topic	Papers
Prompt Injection & Jailbreaks	250
Agent & Tool Security	282
Privacy & Data Leakage	165
Safety, Alignment & Misuse	281
Adversarial ML, Poisoning & Backdoors	252
Software & Vulnerability Security	367
Malware, Phishing & Cyber Defense	144
Evaluation, Benchmarks & Red Teaming	547
Other LLM Security	85

Latest papers

Updated	Paper	Topics	Links
2026-09-08	PrivEscalate: Measuring and Augmenting the Threat of LLM-Automated Linux Privilege Escalation Yixuan Liu, Zilong Zhen, Yin Wu, et al.	Agent & Tool Security, Software & Vulnerability Security, Evaluation, Benchmarks & Red Teaming	abstract / PDF
2026-09-08	AGMark: Attention-Guided Dynamic Watermarking for Large Vision-Language Models Yue Li, Xin Yi, Dongsheng Shi, et al.	Other LLM Security	abstract / PDF
2026-09-08	VectraYX-Vision-1B: A Sub-2B Spanish/LATAM Cybersecurity Vision-Language Model with Structured Visual Reasoning and Native Tool Use Juan S. Santillana	Agent & Tool Security, Malware, Phishing & Cyber Defense, Evaluation, Benchmarks & Red Teaming	abstract / PDF
2026-09-08	Measuring the Security of the Evolving Software Supply Chain: a Research Agenda Sarah Meriem Ourari	Software & Vulnerability Security	abstract / PDF
2026-09-08	Evidence-Grounded Retrieval for Investigation Hunt Lead Generation from CTI Reports Akash Prakash, Boubakr Nour, Makan Pourzandi, et al.	Malware, Phishing & Cyber Defense	abstract / PDF
2026-09-08	Benchmark Scores Are Pipeline-Dependent: A Reliability Audit of Cybersecurity LLM Benchmarks Aymene Berriche, Cathrine Shalby, Mohannad Alhanahnah, et al.	Malware, Phishing & Cyber Defense, Evaluation, Benchmarks & Red Teaming	abstract / PDF
2026-09-08	A Blind Trust, the Bloody Thrust: When Attacker-Controlled Hook Updates Steer AI Agent Harnesses towards Malicious Behaviors Pengxun Li, Litian Zhang, Jianwei Hou, et al.	Software & Vulnerability Security, Evaluation, Benchmarks & Red Teaming	abstract / PDF
2026-09-08	Suan: Rectifying Direct Preference Safety Alignment in Large Language Models Oleksandr Cherednichenko, Roman Klypa	Safety, Alignment & Misuse	abstract / PDF
2026-09-08	Breaking Planner Integrity Boundary: Enviroment State-Text Injection Attack on LLM-Driven Embodied Agents Jiawei Liu, Jiacheng Guo, Tian Zhang, et al.	Other LLM Security	abstract / PDF
2026-09-08	What You See Is Not What AI Gets: DPAgent-in-the-Middle Defense Against AI-Groomed Deceptive Patterns Zewei Shi, Ruoxi Sun, Haoyang Li, et al.	Agent & Tool Security, Privacy & Data Leakage	abstract / PDF
2026-09-08	Structural Jailbreaks Generalize but Do Not Compound: A cross-provider and multilingual study of Involuntary In-Context Learning Tejasvi C. Addagada	Prompt Injection & Jailbreaks, Safety, Alignment & Misuse, Software & Vulnerability Security, Evaluation, Benchmarks & Red Teaming	abstract / PDF
2026-09-08	Do Input-Level Defenses Transfer to Observation-Level Attacks on VideoLLMs? Bangshuo Zhu, Wei Song, Yuxin Cao, et al.	Safety, Alignment & Misuse, Adversarial ML, Poisoning & Backdoors, Evaluation, Benchmarks & Red Teaming	abstract / PDF
2026-09-08	HoneyRoute: Honeypot-Model Routing for Adversarial LLM Serving Han Jin	Other LLM Security	abstract / PDF
2026-09-08	ACEA: An Adversarial Co-Evolution Arena for Head-to-Head Red-Team and Blue-Team LLM Testing Yi Ting Shen, Kentaroh Toyoda, Alex Leung	Evaluation, Benchmarks & Red Teaming	abstract / PDF
2026-09-08	Style Over Substance: Content-Invariant Wrappers Flip LLM Safety-Judge Verdicts Yongxi Zhou, Wenbo Ye, Yuanzhe Liu, et al.	Prompt Injection & Jailbreaks, Safety, Alignment & Misuse, Software & Vulnerability Security, Evaluation, Benchmarks & Red Teaming	abstract / PDF

Scope

Included work must mention an LLM or language-model concept and a concrete security, safety, privacy, abuse, or cyber-defense concept in its title or abstract. The taxonomy covers:

• Prompt injection and jailbreaks
• Agent and tool security
• Privacy, memorization, and data leakage
• Model safety, alignment, and misuse
• Adversarial attacks, poisoning, and backdoors
• Vulnerability discovery and secure software
• Malware, phishing, and threat intelligence
• Security evaluation, benchmarks, and red teaming

The catalog is automated discovery, not a quality ranking or endorsement. See the methodology for the query, scoring rules, known limitations, and correction process.

How it works

arXiv Atom API
      |
      v
query + pagination -> relevance scoring -> revision deduplication
      |                                          |
      +-------------------> data/papers.json <---+
                                  |
                                  v
            README.md + papers.md + CSV + web index

The collector uses only the Python standard library. There is no package installation step and no runtime dependency lockfile to maintain.

# Run tests
python3 -m unittest discover -s tests -v

# Fetch recent papers and regenerate every output
python3 scripts/collect.py

# Regenerate Markdown and CSV without network access
python3 scripts/collect.py --render-only

Search terms and taxonomy rules live in config/topics.json. The canonical record format is documented by data/schema.json. Updates run daily at 06:17 UTC and can also be started manually from the Actions tab.

Data use

• data/papers.json is the canonical, stable dataset.
• data/papers.csv is convenient for spreadsheets and analysis.
• papers.md is the human-readable catalog grouped by topic.
• docs/data contains compact, generated payloads for the SecPapers web index.
• Each record links to the authoritative arXiv abstract and PDF.
• Paper titles, abstracts, and author metadata remain attributable to their respective authors and are not relicensed by this repository's MIT license.

Contributing

False positives, missing papers, taxonomy improvements, and collector fixes are welcome. Read CONTRIBUTING.md before opening a pull request.

Acknowledgments

Paper metadata is provided by the arXiv API. SecPapers is not affiliated with or endorsed by arXiv. Please cite the original authors and papers when using this catalog in research.