Prompt injection: research

Journal articles, academic blogs, think-tank reports and papers in this section, newest first. Back to the news.

arXiv (research papers) Gjergji Kasneci SI

The safety failures we are not instrumenting: a perspective on hidden safety-critical challenges in modern AI systems

Current AI safety discourse still focuses disproportionately on visible failures, including obvious harms, dramatic misuse, and hypothetical catastrophic scenarios. That focus is incomplete. In deployed systems, many of the most consequential failures are quieter: plausible rather than spectacular, distributed across components rather than localized in a single output, and normalized by workflows before they are recognized as hazards. We argue that a central safety challenge…

Read at arXiv (research papers)

arXiv (research papers) Akansha Shukla SI

Data Leakage Prevention in Agentic Applications via Preemptive Hardening

Agentic systems integrate LLM driven planning with interfaces to external tools, making data leakage and tool misuse feasible via instruction/data boundary failures and prompt injection attacks. Enforcing required controls consistently is particularly challenging in workflows spanning many codebases and heterogeneous agents. To address this challenge in multi agentic systems, we present a pre-deployment pipeline for scanning, hardening, and validation of agentic…

Read at arXiv (research papers)

arXiv (research papers) Xinting Liao SI

CPInj: Uncovering Prompt Injection Risks in Textual Collaborative Prompt Optimization

Textual Collaborative Prompt Optimization (TCPO) extends TextGrad (Yuksekgonul et al., 2025) to a decentralized setting by allowing multiple clients to jointly improve prompts for large language models (LLMs) while keeping their data locally. Its reliance on free-form textual updating and aggregation introduces a new and largely unexplored attack surface, i.e., malicious instructions can be injected into local prompts and propagated through server-side prompt aggregation…

Read at arXiv (research papers)

arXiv (research papers) Om Narayan SI

ChainWatch: A Kill Chain-Aligned Sequential Detection Framework for Multi-Step Attacks in MCP-Based AI Agent Systems

The Model Context Protocol (MCP) is an open-source standard that allows AI agents to connect to external tools, databases, and services. While this connectivity enables powerful agent capabilities, it also introduces multi-step attacks that existing per-call defenses cannot reliably detect. Attackers can compose individually benign tool invocations into malicious sequences that evade isolated inspection. This paper presents ChainWatch, a sequential detection framework for…

Read at arXiv (research papers)

arXiv (research papers) Jie Li SI

Trusted Credentials, Untrusted Behavior: Benchmarking LLM-Agent Security in High-Performance Computing

Large language model (LLM) agents are starting to take on routine work in high-performance computing (HPC), including monitoring Slurm jobs, diagnosing failed builds, inspecting simulation output, and coordinating scientific workflows. To do this work, an agent commonly acts under its user's credentials and inherits the user's access to files and the scheduler. This arrangement creates a failure mode that ordinary account-level controls do not capture. Adversarial…

Read at arXiv (research papers)

arXiv (research papers) Devina Jain SI

Adaptive Adversaries: A Multi-Turn, Multi-LLM Benchmark for LLM Agent Security

LLM-based agents process external content, exposing them to prompt injection and multi-turn manipulation. We present a 21-scenario benchmark for adaptive cross-session attacks against fresh-session LLM defenders: an autonomous LLM attacker observes prior defender responses and pivots across rounds, while each defender response is evaluated as a fresh interaction. A controlled 3 x 3 attacker-defender matrix contains 945 battles. Restricting scoring to the first round yields…

Read at arXiv (research papers)

arXiv (research papers) Xingfu Zhou SI

Salience Induction against Multi-Hop RAG Agents: Threat and Defense

Agentic retrieval-augmented generation (RAG) systems increasingly retrieve external evidence and orchestrate tools for knowledge-intensive applications. In Multi-Hop question answering, agents chain facts across documents. Existing defenses focus on content poisoning, which injects false facts, and prompt injection, which embeds directives. We identify a third attack surface: the salience channel, through which fact position, emphasis, framing, and semantic proximity can…

Read at arXiv (research papers)

arXiv (research papers) Haoyan Luo SI

Persistent Sparse Autoencoders: Learning Feature-Specific Timescales in Language Model Representations

Sparse autoencoders (SAEs) decompose language model activations into sparse features, yet these models traditionally encode each token independently, failing to expose information that persists across a sequence. We first show that temporal persistence can naturally emerge in standard SAE features: after a feature activates, the hidden state remains aligned with its direction, and past activations help reconstruct later hidden states. How long this lasts varies widely across…

Read at arXiv (research papers)

arXiv (research papers) Rodrigo Martins Boos SI

Local AI pre-screening for human triple-blind peer review in health sciences

Academic peer review is under mounting strain: NeurIPS 2025 received 21,575 submissions, ICLR 2025 received 11,603, and ICML 2025 received 12,107. This volume has outpaced the supply of qualified reviewers, and large language models (LLMs) are already filling the gap, largely undisclosed. An independent analysis of ICLR 2026 found roughly 21% of its 75,800 peer reviews were fully AI-generated, with over half showing some AI involvement (up from 15.8% in 2024). Documented…

Read at arXiv (research papers)

arXiv (research papers) Yu Cui SI

Refusal is Not Safety! Benchmarking Latent Safety Risks of LLM-Driven Content Humorization

Safety defenses for large language models (LLMs) have been extensively studied, with existing approaches focusing on attack detection and refusal mechanisms. Such fixed-form direct refusal strategies may introduce the risk of prefix injection attacks. Recent work has explored a new direction that leverages humor as an indirect refusal mechanism to mitigate over-refusal in jailbreak scenarios and reduce prefix injection risks. However, this approach implicitly assumes that…

Read at arXiv (research papers)

Embrace The Red Security

From Indirect Prompt Injection to DNS Exfiltration in macOS Terminal

This is a follow-up to my previous Terminal DiLLMa research, and there is a positive outcome: Apple fixed a macOS Terminal behavior that enabled a DNS-based data exfiltration technique. DNS Requests via ANSI Escape Codes David Leadbeater originally discovered an interesting behavior in the macOS Terminal app that allowed a special sequence of ANSI escape codes to issue DNS requests. In short, this triggered a DNS request from the macOS Terminal app:

Read at Embrace The Red

arXiv (research papers) Soham Gadgil SI

Bad Memory: Evaluating Prompt Injection Risks from Memory in Agentic Systems

A growing class of agentic systems maintain persistent state across sessions through memory files, behavioral preferences, and knowledge bases. While this makes agents more useful and self-improving, it also creates a new attack surface for prompt injections in which malicious instructions can be embedded within persistent files and influence future behavior. In this work, we study prompt injection attacks in memory-based agentic systems using a sandboxed synthetic…

Read at arXiv (research papers)

arXiv (research papers) Rabimba Karanjai SI

Context Contamination in LLM Analysis of Network Security Logs: Poison with Passive Prompt Injection and Mitigation Evaluation

Large Language Models are increasingly deployed in Security Operations Centers for log analysis tasks including summarization, alert triage, and threat investigation. These systems ingest logs from external-facing services and process network logs as natural language contexts to generate security insights. We demonstrate that this architectural pattern introduces a critical vulnerability: adversaries can embed prompt injection payloads in log-generating fields that persist…

Read at arXiv (research papers)

arXiv (research papers) Mohammad Allahbakhsh SI

Rethinking Penetration Testing for AI-Enabled Systems: From Resource Compromise to Behavioral Objective Violation

Penetration testing traditionally evaluates whether adversaries can exploit weaknesses in software, infrastructure, configurations, or operational controls to achieve security-relevant compromise. This paradigm remains necessary for AI-enabled systems, but it is no longer sufficient. In such systems, adversaries may influence prompts, retrieved content, sensor inputs, training data, memory, tools, or human-AI interaction loops to alter system behavior without directly…

Read at arXiv (research papers)

arXiv (research papers) Sanket Badhe SI

Agent Skill Security: Threat Models, Attacks, Defenses, and Evaluation

Reusable skills are becoming a fundamental building block of Large Language Model (LLM) agents, enabling capabilities to be packaged, shared, and reused across diverse applications. However, existing security research primarily focuses on prompt injection and runtime execution, leaving security risks throughout the broader skill lifecycle largely unexplored. In this paper, we present SkillSec-Eval, a lifecycle-aware framework for systematically evaluating the security of…

Read at arXiv (research papers)

arXiv (research papers) Junhui Wang SI

PVDetector: Detecting Prompt Injection Attacks on Purpose-Specific LLM Agents through Policy-Violation Concept Analysis

Large language models (LLMs) are increasingly deployed as purpose-specific agents to handle domain-specific tasks such as customer service and code generation. These agents are expected to comply with not only generic safety guardrails but also purpose-specific restrictions tailored to their designated roles. Such additional restrictions enlarge the attack surface, particularly to prompt injection (PI) attacks. To defend against such attacks, existing detection methods…

Read at arXiv (research papers)

arXiv (research papers) Huihao Jing SI

Isolation as a First-Class Principle for LLM-Agent System Safety: Concepts, Taxonomy, Challenges and Future Directions

The capability of LLM agents to function as the ``brain'' of a system fundamentally expands the scope of analysis beyond a standalone model. Consequently, safety is no longer only about input--output content alignment. It also concerns system behavior and real-world execution outcomes. However, the current literature is fragmented across attack types, applications, and benchmarks. This makes it hard to explain why failures such as prompt injection, tool misuse, and memory…

Read at arXiv (research papers)

arXiv (research papers) SingGuard Team SI

SingGuard-NSFA: Extensible Guardrails for Agentic AI via Generative Reasoning and Real-Time Classification

We present nsfaguard, a guardrail framework for securing agentic AI systems against operational threats, such as prompt injection, sensitive information extraction, malicious code requests, dangerous tool misuse, and resource exhaustion. We first introduce the NSFA taxonomy, which organizes 185 risk variants into a CIA-triad-grounded hierarchy and is cross-validated against three well-established OWASP guidelines. Based on this taxonomy, we construct a benchmark suite…

Read at arXiv (research papers)

arXiv (research papers) Bálint Gyevnár SI

Distributed Denial of Science: How Indirect Data Poisoning of AI Systems Can Industrialize Scientific Fraud

Scientific fraud is the instrument of doubt that malicious entities can use to establish controversy in science. Historically, it required the resources of a company: deep pockets, ghostwritten articles, and corrupt academics. Today, Artificial Intelligence (AI) is increasingly automating scientific research, so we ask: Can a remote adversary weaponize the honest use of AI in science to compromise scientific integrity? We envision and empirically evaluate a new attack…

Read at arXiv (research papers)

arXiv (research papers) Ruksat Khan Shayoni SI

NetInjectBench: Benchmarking Indirect Prompt Injection in Tool-Using Large Language Model Agents for Network Operations

Tool-using large language model (LLM) agents are attractive for network operations, but tickets, alerts, logs, runbooks, and ChatOps messages can carry indirect prompt injections. We present NetInjectBench, a 130-scenario benchmark that separates untrusted artifact text, trusted policy metadata, and evaluation labels for network-operation tool use. The sample contains 40 benign, 40 weak-attack, 40 strong-attack, and 10 approved high-impact change scenarios; each is evaluated…

Read at arXiv (research papers)

arXiv (research papers) Yaxin Li SI

Devil in the Lens: Analyzing and Defending Physical Prompt Injection Against Vision-Language Models on Wearable Devices

Vision-Language Models (VLMs) are rapidly deployed on human-facing wearable devices such as smart glasses to enable multimodal perception and AI-assisted decision-making. While prior research has demonstrated the risks of visual prompt injection into digital image inputs of VLMs, the unique security challenges posed by the increasing integration between physical environments and wearable intelligence, such as those embodied in VLM-enabled AI glasses, remain underexplored…

Read at arXiv (research papers)

arXiv (research papers) Nguyen Khoi Vo SI

The Effect of Multi-Lingual and Keyword Adversarial Injection on LLM Relevance Judgment

Large language models (LLMs) are increasingly being used as automated judges for relevance evaluation in information retrieval, yet their robustness to adversarial manipulation remains insufficiently understood, particularly in multilingual settings. In this work, we investigate the impact of cross-lingual prompt injection attacks on LLM-based relevance judgments using TREC Deep Learning collections and two open-weight models under established prompting frameworks. We…

Read at arXiv (research papers)

arXiv (research papers) Zhongzheng Xu SI

"Code Is Cheap. Show Me the Talk.": Lessons from Teaching and Managing AI Coding Tool Usage in a Visualization Course

Generative Artificial Intelligence (GenAI) coding tools are transforming visualization education. They can assist with implementation and design, but they can also let students bypass intended learning trajectories. In this paper, we share our retrospective experience managing and teaching AI use in an upper-level visualization course. We implemented prompt injections, asked oral checkout questions, and taught two AI coding labs. Prior to our coding labs, at least half of…

Read at arXiv (research papers)

arXiv (research papers) Mohammad Dabaja SI

Promptable Concept Segmentation from Above: Evaluating SAM 3's Zero-Shot and One-Shot Capabilities in Remote Sensing

The deployment of large-scale foundation models, such as the Segment Anything Model 3 (SAM 3), promises a transition toward open-vocabulary, training-free computer vision. However, their capacity to generalize out-of-distribution to the complex, top-down geometric structures of Earth Observation imagery remains largely unquantified. Driven by SAM 3's performance disparities in highly specialized domains, we present a comprehensive, multi-task empirical evaluation across…

Read at arXiv (research papers)

arXiv (research papers) Hugo García Cuesta SI

Multi-Agent Firewall Architecture for Privacy Protection of Sensitive Data in Interactions with Language Models

While Large Language Models (LLMs) have become essential productivity tools, their integration into workflows without adequate safeguards creates significant risks. This paper proposes an open-source, privacy-focused, user-facing firewall designed to secure both web-based and programmatic LLM interactions. The architecture combines a browser extension and a proxy for total traffic interception across both HTTP(S) and WebSocket communications. At its core, a flexible…

Read at arXiv (research papers)

arXiv (research papers) Corban Villa SI

Prismata: Confining Cross-Site Prompt Injection in Web Agents

Autonomous web agents promise to automate everyday browsing tasks, but inherit one of the web's oldest attack surfaces. Cross-Site Scripting proved that mixing trusted and untrusted content is dangerous, even on benign pages. Agents resurface this risk by interpreting natural language as instructions, allowing third-party and user-generated content to hijack the agent via prompt injection. The core challenge is that deriving a task-specific security policy requires reasoning…

Read at arXiv (research papers)

arXiv (research papers) Phat T. Tran-Truong SI

FedMark-FM: Auditable, Risk-Adjusted Data Markets for Federated Foundation-Model Adaptation

Federated foundation-model adaptation increasingly relies on heterogeneous private artifacts (retrieval corpora, prompts and demonstrations, LoRA adapters, preference and safety data, and update sketches), yet existing federated-learning incentive mechanisms price clients as homogeneous data or update providers. This assumption poorly matches foundation-model pipelines, where contribution value is heterogeneous, non-IID, pipeline-dependent, privacy-constrained, and…

Read at arXiv (research papers)

arXiv (research papers) Aya Spira SI

Beware of Agentic Botnets: Scalable Untargeted Promptware Attacks via Universal and Transferable Adversarial HalluSquatting

The growing adoption of agentic LLM applications has introduced a new threat previously named as promptware. While prior work has established that adversaries can exploit direct channels to LLM applications to apply promptware under weak threat models, many applications do not provide any direct channels that could be exploited for prompt injection beyond the Internet. This raises a question: can attackers exploit LLM applications at scale without any direct channels in…

Read at arXiv (research papers)

arXiv (research papers) Elad David SI

Latent Undertow: How Ordinary Typos Break Probes

LLMs handle ordinary typing variation fluently: a typo or missing punctuation leaves both user intent and the model's response substantively unchanged. Yet probes that detect malicious prompts by reading the model's hidden states tell a different story: the same edit rotates the readout vector by 43--56 at the perturbed token, decaying below 15% within ~10 downstream tokens. Stacking ~3 common typos per message cuts a single-position prompt-injection probe's TPR@FPR$=1% by…

Read at arXiv (research papers)

arXiv (research papers) Kristina Nikolić SI

Untrusted Content Masking for Web Agents with Security Guarantees

Defenses that provide security guarantees against prompt injection attacks rely on strict isolation between trusted instructions and untrusted data. In text-based environments such as tool-use APIs, this separation arises naturally: agents can reason from interface definitions without ever processing untrusted content. Extending these guarantees to web agents faces a fundamental challenge: to perceive and interact with their environment, web agents must first observe the…

Read at arXiv (research papers)

Older stories