Prompt injection: research

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

arXiv (research papers) Riya Bharathwaj SI

ErosionBench: Probing Alignment Decay in Enterprise Multi-Agent Systems

As enterprises delegate consequential work to multi-agent LLM systems, benign-looking requests can escalate into unsafe actions as they propagate across sessions through poisoned shared state, long-context documents, and trusted peer messages: no single message is malicious, yet the assembled tool action is. Prior benchmarks isolate a single agent under prompt injection or score multi-agent attacks by their final output, leaving cross-session erosion of shared state out of…

Read at arXiv (research papers)

arXiv (research papers) Bibek Bhandari SI

Do Benign-Trained Hidden-State Probes Detect Tool Errors Under Adversarial Conditions?

Although hidden-state probes efficiently monitor model operation, it remains unclear whether probes trained on benign tool use remain useful for detecting incorrect actions under adversarial conditions. In this research, we assess this problem from the perspective of operational tool-action correct- ness. we use four AgentDojo suites, four conditions, and three models: Qwen2.5-7B-Instruct, Qwen2.5-14B-Instruct, and GLM- 4.6V-Flash (9B). We perform a five-fold, task-grouped…

Read at arXiv (research papers)

arXiv (research papers) Thanh M. Truong-Le SI

Agents of Chaos: A Systematization of Knowledge (SoK) on Malicious Attacks on Agentic AI Systems

Agentic AI systems extend large language models with tools, memory, retrieval, browser or computer-use interfaces, code execution, external services, and multi-agent coordination. Although these capabilities enable agents to act beyond text generation and perform tasks, they also create attack paths through which adversaries can exploit untrusted content, deceptive metadata, poisoned state, compromised skills, mutable dependencies, or malicious code. Such attacks can…

Read at arXiv (research papers)

arXiv (research papers) Yichen He SI

CompositionEvidence: Controlling Geometric Evidence for Cinematic Composition Reasoning with Frozen Vision-Language Models

Cinematic composition questions depend on in-frame geometry that vision-language models (VLMs) use inconsistently. External perception tools can expose this geometry, but naive prompt injection can overwrite useful visual predictions. On ShotBench composition, pre-question geometry reduces Qwen2.5-VL-7B from 49.5% to 46.3%; corrupted measurements lower it to 33.6%. CompositionEvidence derives nine reproducible features from frozen detection, depth, and line experts and…

Read at arXiv (research papers)

arXiv (research papers) Mohammed Nayeem Khan SI

Enmo Safety: External, Non-invasive, Model-agnostic Oversight for Off-the-Shelf Language Models

We study a safety harness that improves itself without touching model weights. Enmo Safety wraps a chat model with a deterministic input assessment, trigger-matched safety guidance, spotlighting of untrusted content, and an output guard. A reflector model turns failures into one bounded update per round, and an update is kept only if it lowers harm without raising refusal of benign prompts on a held-out dev split. Every version is snapshotted and every decision ledgered. We…

Read at arXiv (research papers)

arXiv (research papers) Mark Heimann SI

Notwithstanding Anything to the Contrary, Ignore All Instructions: Measuring Prompt Injection Guards on Legal Contract Review Attacks

Specialized prompt injection guards are deployed in front of document analysis systems, but no false positive rate has been published for any of them on legal contracts. We measure eleven open-weights guards on two corpora of legal contracts and find that existing guards struggle: document-level false positive rates are as high as 92.39%; two guards fall below chance on paired AUC; and a five-regex baseline is not Pareto-dominated by any guard in the panel. Controlled probes…

Read at arXiv (research papers)

arXiv (research papers) Parth Illendula SI

I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents

Large Reasoning Models (LRMs) are one of the key components of current agentic systems. LRMs augment their output with an explicit chain-of-thought reasoning stage, introducing a new attack surface for thought injection, in which untrusted text resembling the model's own deliberation is fed back to the model. We present a two-part study of this attack, performed against the prevailing large, production class open-weight models (e.g., GLM-5.3, MiniMax-M2.5…

Read at arXiv (research papers)

arXiv (research papers) Julian Kunkel SI

The Power of Meta Access in the Software of the Future --- Demonstrated on LLM Chat UIs and Chat AI

Chat remains the default surface for many large language model (LLM) tasks, even as agentic tools such as OpenCode excel at coding and automation. Yet the model sees the current dialogue but not the workspace around it. We explore meta access---giving the assistant a structured, permissioned view of the interface itself and propose a ladder of levels: feature tutoring; metadata, settings, and preference access; retrieval over summaries and messages, confirmed mutation, and…

Read at arXiv (research papers)

arXiv (research papers) Rongting Zhang SI

Beyond Instruction Override: Benchmarking Indirect Prompt Injection with Task-Tampering Attacks

Indirect prompt-injection (IPI) benchmarks predominantly test planted instructions that redirect a tool-using agent toward an attacker-chosen task. Yet attacker-controlled content can instead tamper with the outcome of the task already in progress. A buyer's dispute evidence can cite a false refund policy to inflate a refund without changing the task. We define IPI by authority rather than instruction-like wording: planted content constitutes an attack when it attempts to…

Read at arXiv (research papers)

arXiv (research papers) Peiran Wang SI

MetaPI: Constructing Prompt Injection Benchmarks from LLM Agent Benchmark

Large language model (LLM) agents increasingly perform complex tasks using high-privilege tools, making indirect prompt injection (IPI) a critical security risk. However, IPI benchmarks lag behind their rapidly evolving utility counterparts, only offering limited task and environment coverage and relying on narrow, static attack scenarios. In this paper, we propose meta-benchmarking for IPI: transforming existing utility benchmarks into end-to-end IPI evaluation suites that…

Read at arXiv (research papers)

arXiv (research papers) Woosang Lim SAINT Machines KAIST SI

SPILLWAY: Separating Target Admission from Sanitization to Defend Tool-Using Agents against Prompt Injection

Tool-using agents act on email, files, and payments, so an instruction hidden in a web page or a shared document can make an agent send data or money to an attacker. Many runtime defenses put a language model on the decision that lets an action through, so the decision depends on a model that may read the attacker's text and adds model calls to every step. Provenance defenses keep the model off the final check but need a model to plan or route each task first. Where an…

Read at arXiv (research papers)

arXiv (research papers) Guoqin Xiong SI

SecureGARD: Dynamic Ontology-Guided Governance for Secure Tool-Using LLM Agents

Tool-using large language model (LLM) agents translate model outputs into consequential actions, turning indirect prompt injection into an execution-security problem. Existing defenses constrain tool calls, trajectories, or information flows, but often leave action semantics, state dependencies, and evidence relations implicit, limiting generalization across heterogeneous tools and complicating safe updates. We present SecureGARD, a dynamic ontology-guided governance…

Read at arXiv (research papers)

arXiv (research papers) Annabella Chow SI

Securing Web Agents with the Compile-Then-Execute Paradigm

Web agents commonly follow the ReAct paradigm, selecting each action from runtime observations that inherently mix trusted and untrusted content. This design makes their control flow vulnerable to prompt injections embedded in webpages. We introduce Baya, a compile-then-execute web agent designed to prevent prompt-injection-driven control-flow hijacking by design. Before accessing the live website, Baya compiles the user's request into an executable Python program that calls…

Read at arXiv (research papers)

arXiv (research papers) Yutao Wu SI

ScamWorld: An Interactive Environment for Evaluating Web Agents Against Online Fraud

Web agents increasingly browse, shop, and manage accounts on behalf of their users. Prior work has shown that they can be misled by adversarial content, including prompt injections and deceptive interfaces. Beyond such attacks, agents operating on the open web also encounter online fraud designed to deceive humans, including fraudulent online stores, fake investment platforms, and advance-fee schemes. How web agents respond to such fraud during task execution remains…

Read at arXiv (research papers)

arXiv (research papers) Yao Wang SI

Right Tool, Wrong Parameters: Defending LLM Agents Against Parameter Poisoning

Indirect Prompt Injection (IPI) poses a critical threat to LLM agents. Existing defenses largely focus on whether an injection induces a new tool invocation or unintended action. In this paper, we study a different attack surface, parameter poisoning, where the attacker preserves the intended action but manipulates one or more tool arguments, such as a recipient or other task-critical value. Such attacks are difficult to detect because no new action is introduced, while…

Read at arXiv (research papers)

arXiv (research papers) Jiamin Chang SI

The Devil is in the Context: Nine Dimensions of Prompt Injection Attacks

Prompt injection is usually discussed and evaluated as a single, homogeneous type of attack. However, vulnerabilities reported in real-world systems show that prompt injections vary widely. Moreover, the same instruction can be a prompt injection in one context and a legitimate request in another, depending on the user's intent and the system's policies. Ignoring this variation leaves gaps in how defenses are evaluated, and can give a false sense of security against…

Read at arXiv (research papers)

arXiv (research papers) Gia Phong Tran SI

From Guidelines to Practice: Defining Requirements for AI Red-Teaming Tools

Artificial Intelligence (AI) has seen significant developments and increased usage in recent years. AI benefits various tasks, automates decision-making, and enhances efficiency. However, AI also introduces additional attack surfaces for threat actors to perform attacks like prompt injection and jailbreaking, making security a great concern. AI security testing aims to imitate adversarial attacks or malicious inputs via manual testing or with the assistance of tools. Thus…

Read at arXiv (research papers)

arXiv (research papers) Xu Zhang SI

ATrust: Adaptive Trust Boundary Expansion for Securing LLM Agents

As large language model (LLM) agents take on more complex tasks with greater autonomy, security has become a major obstacle to their broader deployment. Unlike in attacks that require complex exploitation of operating system vulnerabilities, attackers can easily hijack agents through indirect prompt injection (IPI), leading to privacy leakage, financial loss, and destructive operations. However, existing defenses often make security decisions based on limited static trusted…

Read at arXiv (research papers)

arXiv (research papers) Kainan Zhou SI

SearchInject: Retrieval Exposure Control for Indirect Prompt Injection in Language-Model Search

An indirect prompt injection can affect a search-augmented large language model only after its carrier page reaches model context. SearchInject controls this admission decision with a joint score combining retrieval relevance, search-abuse cues, instruction likelihood, and two relevance interactions. A quarantine threshold is selected under an empirical 0.5% clean-candidate false-quarantine constraint on a calibration split. Evaluation uses 300 held-out SciFact queries and…

Read at arXiv (research papers)

arXiv (research papers) Christopher Krakauer SI

Where Does MCP Security End? A Three-Tier Mitigation Mapping Framework for the OWASP MCP Top 10

The Model Context Protocol (MCP) has become a leading standard for connecting Large Language Models (LLMs) to external tools, software, and data sources. MCP architecture favors ease of integration over security, blending the data and control planes into a single natural language channel. This introduces confused deputy and prompt injection vulnerabilities that cannot be completely mitigated in the software development lifecycle (SDLC) of an MCP server. Structured security…

Read at arXiv (research papers)

arXiv (research papers) Haowen Dai SI

StateFence: Task-State Protection Against Evidence-Chain Contamination in Blue-Team LLM Agents

Indirect prompt injection creates a distinctive risk for blue-team Large Language Model (LLM) agents in Security Operations Centers (SOCs). Malicious instructions can be embedded in telemetry, tickets, retrieval results, or memory and then propagate through alert triage, incident summarization, response recommendation, and Indicator of Compromise (IOC) extraction as locally plausible intermediate decisions. We characterize this failure mode as evidence-chain contamination of…

Read at arXiv (research papers)

arXiv (research papers) Junwei Yao SI

Ranked but Not Stopped: Detecting Prompt-Injection Hijacks from Typed Decision Models' Output Probabilities

Programs that use a language model as a classifier act on its most probable label. Text injected into the input can move that label to one the attacker chooses, a hijack. A monitor that does not know the attacker's label can read the model's own confidence and raise an alarm when the top probability is low. We test this monitor on AG News topic classification with five systems, three of them served through a provider's API. Pooled over five families of injected text, low…

Read at arXiv (research papers)

arXiv (research papers) Javad Mohammad Alizadeh SI

Benchmark of THReD: Evaluating an LLM Agent and Its Preprocessing Tools for Turning Electronic Health Records into AI-Ready Datasets

LLM agents are increasingly built into data-analysis software, yet little evidence shows how reliably they preprocess clinical data or whether the tools they invoke give correct, reproducible output. We report a benchmark of THReD, a web-based system that turns electronic health record (EHR) extracts into AI-ready datasets through a chat agent whose proposed steps run only after user confirmation. The agent benchmark has 141 tasks (100 templates) in seven categories…

Read at arXiv (research papers)

arXiv (research papers) Sreehari Sankar SI

LEANGUARD: Lean-Verified Guardrails for Tool-Using Agents

Tool-using language-model agents often execute actions that their domain policies forbid. We present LeanGuard, a runtime that checks every proposed tool call against a policy written in Lean4 before the call reaches the environment. Policies combine state predicates with scoped past-time temporal operators over the execution history, evaluation is fail-closed, and the decision procedure is proved equivalent to a LeanLTL semantics. A denied call can be corrected in two ways…

Read at arXiv (research papers)

arXiv (research papers) Shiyu Kan Shanghai Jiao Tong University, Shanghai, China SI

Beyond Prompt Injection: The Tool Update Gap in LLM Agent Security

Large language model based agents increasingly rely on external tools to invoke services and interact with web resources. These tools can evolve through software updates, yet agentic frameworks may continue trusting them without re-verifying their safety after implementation changes. This tool update gap creates a supply-chain vulnerability in agentic systems: because updated tool code executes outside the LLM's context window, malicious functionality embedded in tool…

Read at arXiv (research papers)

arXiv (research papers) Zhihua Wang SI

MemHarmBench: Evaluating Cross-Session Prompt Injection in Memory-Persistent AI Agents

AI agents often rely on persistent memory to retain user preferences and operational rules across sessions. However, this capability introduces a new safety risk: malicious instructions from messages, documents, or Web content can be retained in memory and influence future tasks even after the original attack content is removed. Existing attack success metrics, however, fail to capture who is affected and how attacks translate into downstream harm during otherwise normal…

Read at arXiv (research papers)

arXiv (research papers) Aravinda Kashyap Kabbinamakki Suresh SI

AgentCap: Capability-Based Runtime Authorization for Secure LLM Agent Execution

Language-model agents commonly collapse two decisions that security systems separate: selecting an action and determining whether that action is authorized. When a model exercises standing authority, indirect prompt injection can convert influence over its reasoning into unauthorized effects. We present AgentCap, a runtime authorization layer that externalizes this decision from the model and mediates every tool invocation through an MCP wrapper. Each request declares an…

Read at arXiv (research papers)

arXiv (research papers) Xingling Xu SI

The Long Con on the Web: Earn-then-Exploit Fraud Attacks against Coding Agents

Coding agents increasingly operate with web access and high privileges, creating a new attack surface where web fraud can directly target autonomous agents. Existing studies mainly focus on malicious links, prompt injection, and interface deception, while content-level web fraud against agents and the amplification of attacks through agent trust remain underexplored. In this paper, we present EtE (Earn-then-Exploit), a web fraud attack framework against coding agents, in…

Read at arXiv (research papers)

arXiv (research papers) Sizhe Chen SI

SecPO: Principled Adversarial Training for Prompt Injection Security

Recognized as a top threat to AI agents, prompt injection allows instructions from untrusted data (email, documents, websites) to manipulate the agent's operation. The existing best-defended LLM, Meta-SecAlign, is trained with the DPO loss but still suffers from attack success rates (ASRs) approaching 100% under adaptive prompt injections. One reason is that DPO anchors the model around the undefended reference model, which is insecure. We show an idealized reference…

Read at arXiv (research papers)

arXiv (research papers) Shuo Hong SI

Erase the Injected Region: Multimodal Prompt-Injection Defense with On-Policy Self-Distillation

Multimodal indirect prompt injection embeds malicious instructions in external media that also contains information needed for the user's task. Existing defenses can block the attack yet impair the model's use of that information, leaving a gap between attack defense and task preservation. We propose ERASE, an on-policy self-distillation method that constructs its reference by erasing the injected region in place. Modality-appropriate blank content replaces the injection…

Read at arXiv (research papers)

Older stories