AI Threat AlertAttackMEDIUM
Adversarial Entropy Inflation Against Gumbel-Based Inference Verification
Nikita Kezins
Gumbel-based inference verification bounds LLM weight exfiltration by only forgiving token choices that plausibly arise from honest GPU...
AI Security Research
AI Threat Alert indexes 4,319+ peer-reviewed and preprint papers on AI/ML security — covering adversarial attacks, model defenses, red-teaming benchmarks, surveys, and security tooling. Papers are sourced from arXiv, classified by type and by relevance to real-world threats, and cross-referenced with the CVEs and incidents they relate to.
- Adversarial attacks
- Model defenses
- Red-teaming benchmarks
- Surveys
- Security tooling
- Total4,319
- Attack1,676
- Benchmark1,229
- Defense610
- Tool432
- Survey252
Showing 141–160 of 609 papers
Clear filtersAttackMEDIUM
Walking on the DARKSIDE
Aldo Gangemi, Emanuele Bottazzi
Large Language Models (LLMs) recognise patterns but do not natively track the path of exclusions that a coherent discourse demands. When an input...
AttackMEDIUM
Breakout/Interchange Reconnection as a driver of Jets, Fast CME, and Solar Energetic Particles
Pankaj Kumar, Judith T. Karpen, David Lario +4 more
Understanding how energetic particles are accelerated and released from the low corona into the interplanetary medium during solar eruptions is...
AttackMEDIUM
Model-Consistent Byzantine-Resilient Decentralized Federated Learning for Collaborative Missions
Yue Li, Sudip Bhujel, Cameron Lira +2 more
Decentralized federated learning (DFL) is a promising paradigm for autonomous nodes to collaboratively train AI models without relying on a central...
AttackMEDIUM
QUASAR: A Quantum-Classical Neural Network for SAR Satellite Physical-Layer Authentication
Vincenzo Sammartino, Nathanael Denis, Roberto Di Pietro
X-band SAR satellites (8-12 GHz) play a critical role in disaster response, environmental monitoring, and military intelligence. Yet, they lack...
AttackMEDIUM
Inadvertent Context Leakage in Language Models
Jaiden Fairoze, Neal Mangaokar, Kamalika Chaudhuri +2 more
For AI agents to be useful beyond simple chat, they must hold sensitive user context such as calendars, credentials, health records, and financial...
AttackMEDIUM
Breaking the weakest link to evade vision language models
Ilan Zini, Boussad Addad, Katarzyna Kapusta
Vision Language Models (VLMs) have recently emerged as a critical component of multimodal AI systems, enabling joint reasoning over visual and...
AttackMEDIUM
FedLNS: Leverage LayerNorm Signature Modeling to Mitigate Adversarial Manipulation in Federated LLMs
Kai Li, Jong-Ik Park, Carlee Joe-Wong +2 more
Federated training enables language models to learn from distributed private text, but the server cannot directly verify the local supervision or...
AttackMEDIUM
Towards Safer RAG: Only Agents Capable of System 2 Thinking may Access Untrusted Documents
Mehrdad Ghassabi
Retrieval-Augmented Generation (RAG) has significantly enhanced the performance of large language models (LLMs), yet these systems remain vulnerable...
AttackMEDIUM
SkillWatermark: An Embedded Skill Watermark of Progressive Privacy Inference via Benign Prompts
Yu Li, Liqi Zhuang, Dong Wei +5 more
Skills for large language model (LLM) agents have been widely deployed across diverse application domains. However, we observe that these skills...
AttackMEDIUM
Conjunctive Poisoning in AI Supply-Chain Applications
Nokimul Hasan Arif, Qian Lou, Mengxin Zheng
Large Language and Vision-Language Models are increasingly deployed through inference pipelines that include prompt wrappers (e.g., templates and...
AttackMEDIUM
Convergent Detour Hijacking: Task-Preserving Resource Amplification in Skill-Based LLM Agents
Junliang Liu, Ruoyu Li, Wenxin Tang +4 more
LLM agents increasingly rely on third-party skills, using natural-language descriptions for selection and instruction bodies for planning. This...
AttackMEDIUM
Rethinking Agent Security as a Networking Problem
Van Tran, Taveesh Sharma, Tajveer Singh Dhesi +1 more
AI agents are rapidly becoming more capable and widely deployed, promising substantial gains in productivity and enabling new classes of...
AttackMEDIUM
Understanding Content Moderation in Large Language Models through Restricted Books: From Refusal to Warning
Xucheng Yu, Emily Knox, Haohan Wang
As large language models enter everyday information pipelines, understanding how they handle sensitive topics matters as much as understanding...
AttackMEDIUM
Learning to Persuade Exposes How Easily LLMs Abandon Correct Beliefs
Nimet Beyza Bozdag, Emre Can Acikgoz, Gokhan Tur +1 more
Persuasion is a core dynamic of natural language communication, shaping how large language models (LLMs) update beliefs, resolve disagreements, and...
AttackMEDIUM
Defending against Model Extraction for GNNs with Model Reprogramming
Yan Wen, Zhenyi Wang, Heng Huang
Graph Neural Networks (GNNs) serve as the backbone for high-stakes applications in Machine-Learning-as-a-Service (MLaaS). Still, their black-box...
AttackMEDIUM
Once Poisoned, Arbitrarily Controlled: A Programmable Backdoor in VLMs
Tao Lin, Gaojie Jin, Zongxin Liu +2 more
Existing vision-language model (VLM) backdoors are usually treated as static vulnerabilities: one-to-one and N-to-N attacks bind one or more triggers...
AttackMEDIUM
Trigger the Straggler: Load Hijack on Mixture-of-Experts LLMs
Rui Zhang, Wenbo Jiang, Hongwei Li +5 more
Expert parallelism (EP) is a common strategy for serving large Mixture-of-Experts (MoE) models across multiple GPUs by distributing experts among...
AttackMEDIUM
Measuring Semantic Abstractness of SAE Features via Nonlocality
Chuqiao Lin, Shivaji Sondhi, Xiao-Liang Qi
Sparse autoencoders (SAEs) have helped uncover mechanistic explanations for LLM behaviours such as reasoning, jailbreaking etc., via understanding...
AttackMEDIUM
Stealing Reasoning Traces from Proprietary LLM APIs
Alexander Panfilov, David Schmotz, Ilia Shumailov +5 more
Leading large language model providers now conceal their models' step-by-step reasoning, or chain-of-thought, to protect intellectual property and...
Frequently asked questions
What is AI security research?
AI security research studies how AI and machine-learning systems can be attacked and defended — covering adversarial examples, prompt injection, model poisoning, training-data extraction, and the mitigations against them. AI Threat Alert curates this research from academic sources so security teams can track the threats behind emerging AI risks.
How many AI security papers does AI Threat Alert track?
AI Threat Alert indexes 4,319+ papers on AI/ML security, classified across attack, defense, benchmark, survey, and tool categories and updated continuously.
Where do the research papers come from?
Papers are sourced from arXiv, then classified by type and by relevance to real-world AI/ML threats, and cross-referenced with the CVEs and incidents they relate to.
What topics does the AI security research cover?
Coverage spans adversarial attacks, model and system defenses, red-teaming benchmarks, literature surveys, and security tooling for LLMs, ML libraries, AI agents, and inference pipelines.
How is this different from a generic paper search?
Every paper is filtered for AI security relevance and linked to the vulnerabilities, vendors, and incidents it relates to, so the research connects directly to operational threat intelligence.
Track AI security vulnerabilities in real time
Get breaking CVE alerts, compliance reports (ISO 42001, EU AI Act), and CISO risk assessments for your AI/ML stack.
Start 14-Day Free Trial