Global ETD Search

Search theses and dissertations gathered from participating repositories worldwide. Every result links back to the library that holds it. No account is needed.

Results

Showing 1 to 20 of 25 for “"Prompt Injection"”.

  1. InjectBench: An Indirect Prompt Injection Benchmarking Framework

    … it also introduces the risk of indirect prompt injection attacks. In such scenarios, an attacker embeds malicious instructions within the retrieved third party data, which when processed by the LLM, can generate harmful and untruthful outputs for an unsuspecting user. Although previous …

    vt Repository record for InjectBench: An Indirect Prompt Injection Benchmarking Framework (opens in a new tab)

  2. Architectural Defenses for Agentic Systems: Prompt Injection and Beyond

    … emerge and stand out from this interaction: prompt injection and supply chain attacks. We propose architectural and deterministic defenses for these two problems. Firstly, we present a taxonomy of tasks based on the minimum data knowledge required for task completion by agents. Motivated by …

    vt Repository record for Architectural Defenses for Agentic Systems: Prompt Injection and Beyond (opens in a new tab)

  3. GuardRAG: Adversarial Preference Training Against Indirect Prompt Injection Attacks

    … risks through the retrieval pipeline. Indirect Prompt Injections (IPIs) occur when hidden instructions inside retrieved documents enter a model’s context as trusted evidence, exploiting weak guardrails and leading to unauthorized data leaks, policy overrides, or attacker-controlled behavior. …

    washington Repository record for GuardRAG: Adversarial Preference Training Against Indirect Prompt Injection Attacks (opens in a new tab)

  4. Optimized Multi-Agent Defense Pipeline Against Advanced Prompt Injection Attacks

    … in production remain critically vulnerable to prompt-injection attacks. These attacks embed adversarial inputs that hijack the model into leaking data, taking unauthorized actions, or producing harmful content. Existing multi-agent defenses handle static, known attacks well but leave three …

    washington Repository record for Optimized Multi-Agent Defense Pipeline Against Advanced Prompt Injection Attacks (opens in a new tab)

  5. Analýza rizik prompt injection útoků v aplikacích využívajících velké jazykové modely

    Útoky typu prompt injection predstavujú rastúcu bezpečnostnú hrozbu pre aplikácie založené na veľkých jazykových modeloch (LLM). Napriek svojej jednoduchosti môžu tieto útoky výrazne ovplyvniť správanie modelu a potenciálne viesť k odhaleniu citlivých informácií alebo k neúmyselným akciám systému. …

    brno-tech Repository record for Analýza rizik prompt injection útoků v aplikacích využívajících velké jazykové modely (opens in a new tab)

  6. Instruction-Hierarchy Violations and Robustness of Large Language Models to Prompt Injection Attacks

    … of modern Large Language Models (LLMs) against prompt injection attacks. Prompt injection is formalized as a violation of instruc tion hierarchy, wherein untrusted user-provided input overrides, modi es, or circumvents trusted system-level instructions. Rather than relying on informal …

    concordia Download PDF for Instruction-Hierarchy Violations and Robustness of Large Language Models to Prompt Injection Attacks (opens in a new tab)

  7. A MIDDLEWARE SYSTEM FOR DETECTING AND BLOCKING MALICIOUS PROMPT INJECTION ATTACKS IN LLM-BASED APPLICATIONS

    Prompt injections can redirect a large language model (LLM) from an application’s intended instructions. In this project, a local middleware system was developed that checks user prompts before passing them on to the model that generates responses. Three detection methods were compared: a …

    csusb Repository record for A MIDDLEWARE SYSTEM FOR DETECTING AND BLOCKING MALICIOUS PROMPT INJECTION ATTACKS IN LLM-BASED APPLICATIONS (opens in a new tab)

  8. Towards Private and Trustworthy Generative Models

    … open-generation settings. Next, we present Hard Prompts Made Easy (PEZ), a gradient-based discrete optimization framework that automates the discovery of adversarial prompts, revealing vulnerabilities in safety-aligned models. This work provides a systematic framework for auditing content filters …

    maryland Repository record for Towards Private and Trustworthy Generative Models (opens in a new tab)

  9. Vectorising Tabular Data for RAG and Vector Database Validation - A study of tabular data retrieval and anomaly detection methods against poisoning attacks in RAG

    … poisoning attacks. The two attacks tested are prompt injections and hate speech. Four table representation methods are evaluated on a corpus of 23,638 chunks from Swedish government and municipality documents. The methods are the following: Unmodified Tabular Extraction (UTE), Metadata-Enriched …

    chalmers Repository record for Vectorising Tabular Data for RAG and Vector Database Validation - A study of tabular data retrieval and anomaly detection methods against poisoning attacks in RAG (opens in a new tab)

  10. Implementace a evaluace útokových modulů pro LLM red-teaming framework

    … ověřovat jejich odolnost vůči útokům, jako jsou prompt injection, obcházení bezpečnostních filtrů nebo extrakce dat. Navržené řešení podporuje různorodé typy útoků, přičemž si zachovává konfigurovatelnost a rozšiřitelnost. Na základě tohoto návrhu bylo vytvořeno 28 útokových modulů adaptovaných z …

    brno-tech Repository record for Implementace a evaluace útokových modulů pro LLM red-teaming framework (opens in a new tab)

  11. Security and Privacy of Machine Learning Systems

    … agentic systems, where the central threat is prompt injection: text in the data an agent reads that is mistaken for instructions. We first build AgentDojo, a dynamic environment that evaluates a tool-calling agent as a whole, and find that even state-of-the-art agents are vulnerable. We then …

    ethz Repository record for Security and Privacy of Machine Learning Systems (opens in a new tab)

  12. Trustworthy Intelligence Fusion for Search and Rescue: Designing Grounded and Secure Multi-Agent AI for Mission Decision Support

    … examines indirect adversarial threats, including prompt injection, knowledge-base poisoning, and adversarial examples, and evaluates corresponding lightweight defences. The results show that modular specialization, probabilistic grounding, and adversarial defences can improve the reliability and …

    calpoly Repository record for Trustworthy Intelligence Fusion for Search and Rescue: Designing Grounded and Secure Multi-Agent AI for Mission Decision Support (opens in a new tab)

  13. Lamb: An advanced architecture for secure agents - Extending agentic AI utility while maintaining strong security guarantees through information flow control

    Indirect prompt injection attacks are a known and severe problem for the emerging technology of agentic AI. To mitigate them, existing deterministic agents rely on the Dual LLM pattern that disallows untrusted sources from influencing control flow. The pattern prevents execution of benign subtasks …

    chalmers Repository record for Lamb: An advanced architecture for secure agents - Extending agentic AI utility while maintaining strong security guarantees through information flow control (opens in a new tab)

  14. A Privacy-Preserving, AI-Driven Framework For Bathroom Activity Monitoring And Health Recommendation In Elderly Care

    … framework was implemented to protect against prompt injection and jailbreaks. This work demonstrates that secure, autonomous health monitoring is achievable using commodity hardware like the CarePredict Tempo device, offering viable pathway for aging-in-place technologies.

    nodak Repository record for A Privacy-Preserving, AI-Driven Framework For Bathroom Activity Monitoring And Health Recommendation In Elderly Care (opens in a new tab)

  15. SCALES: Dual Information-Theoretic Approaches to Mitigating and Quantifying Prompt Injections in RAG Pipelines

    … (RAG) systems are vulnerable to indirect prompt injection, where adversarial instructions embedded in retrieved documents cause the model to override its intended behaviour, exfiltrate confidential context, or execute unauthorised actions. This thesis makes two complementary contributions …

    washington Repository record for SCALES: Dual Information-Theoretic Approaches to Mitigating and Quantifying Prompt Injections in RAG Pipelines (opens in a new tab)

  16. Mitigating Prompt Injections in LLM Applications: A Detection and Filtering Framework

    … system that identifies a wider range of prompt-injection patterns that traditional rule-based systems may miss. This was accomplished by designing a multi-layer detection architecture consisting of four components: banned-word matching, regex pattern detection, an LLM-based contextual …

    eastern-wash Repository record for Mitigating Prompt Injections in LLM Applications: A Detection and Filtering Framework (opens in a new tab)

  17. Artificial Intelligence Strategies for Advancing Global Bioinformatics Infrastructure and Ecosystems

    … on the OpenAI GPT Store are vulnerable to prompt injection, leaking sensitive data and private configuration details. To improve the findability and usability of open-source bioinformatics tools, we introduce ToolsyBio, a modular system leveraging retrieval-augmented generation (RAG) to …

    penn Repository record for Artificial Intelligence Strategies for Advancing Global Bioinformatics Infrastructure and Ecosystems (opens in a new tab)

  18. Towards Understanding Security and Privacy Risks in the LLM Lifecycle

    … inference-time risks through manually crafted prompt injection queries, or assessing training-time data privacy using traditional membership inference attacks (MIAs). However, manual prompt leaking approaches suffer from inefficiency when attacking closed-box applications. Moreover, existing …

    jhu Repository record for Towards Understanding Security and Privacy Risks in the LLM Lifecycle (opens in a new tab) Embargoed until 2028-08-01

  19. On the Release-Readiness of ML-Based Products

    … and defense against adversarial attacks like prompt injection. This thesis explores release-readiness across traditional and generative AI systems. Traditional approaches often follow established checklists and assume newer models are superior, disregarding the utility of older models …

    queens Repository record for On the Release-Readiness of ML-Based Products (opens in a new tab)

  20. From Adversarial Machine Learning to AI Security Risk: Applying Cybersecurity Methods to Artificial Intelligence

    … Score to compare AI security coverage, using prompt injection and model evasion as case studies. Building on these findings, the thesis proposes an AI Security Fundamentals Framework for industry application. Collectively, the research advances AI security as an interdisciplinary field …

    unsw Repository record for From Adversarial Machine Learning to AI Security Risk: Applying Cybersecurity Methods to Artificial Intelligence (opens in a new tab)

Page 1 of 2