Agent Runtime Security 固定订阅页

《Jailbreaking for the Average Jane: Choosing Optimal Jailbreaks via Bandit Algorithms for Automatically Enhanced Queries》〔评测 / 应用 / 方法〕：With a profusion of jailbreaks for LLMs now widely known, a growing concern is that…

1. Jailbreaking for the Average Jane: Choosing Optimal Jailbreaks via Bandit Algorithms for Automatically Enhanced Queries · Score 64
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. Do Safety Guardrails Need to Reason? LeanGuard: A Fast and Light Approach for Robust Moderation · Score 59
   title matched "guardrail"；has PDF；has rich summary
   原始来源
3. AgentX: Towards Agent-Driven Self-Iteration of Industrial Recommender Systems · Score 41
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
4. MIRROR: Novelty-Constrained Memory-Guided MCTS Red-Teaming for Agentic RAG · Score 40
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《How Reliable Is Your Jailbreak Judge? Calibration and Adversarial Robustness of Automated ASR Scoring》〔方法〕：Almost every paper on LLM jailbreaks and prompt injection reports an attack-success rate (ASR), and that number…

1. How Reliable Is Your Jailbreak Judge? Calibration and Adversarial Robustness of Automated ASR Scoring · Score 78
   title matched "jailbreak"；summary matched "prompt injection"；has PDF
   原始来源
2. The Unfireable Safety Kernel: Execution-Time AI Alignment for AI Agents and Other Escapable AI Systems · Score 48
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
3. AI Snitches Get Glitches: Towards Evading Agentic Surveillance · Score 44
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《Burnyard: Future of Malware Analysis》〔方法〕：Malware analysis is a critical aspect of modern cybersecurity. The prevailing industry practice, sandboxing, involves executing suspicious binaries within isol…；《LLMs Prompted…

1. Burnyard: Future of Malware Analysis · Score 47
   summary matched "sandboxing"；has PDF；has rich summary
   原始来源
2. LLMs Prompted for Legal Context Object More: Overrefusal from Small On-Premises LLMs in Criminal Legal Context · Score 44
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
3. Red-Teaming the Agentic Red-Team · Score 43
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
4. PHANTOM: A Large-Scale Dataset of Multimodal Adversarial Attacks for Vision-Language Models · Score 41
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
5. Securing LLM-Agent Long-Term Memory Against Poisoning: Non-Malleable, Origin-Bound Authority with Machine-Checked Guarantees · Score 39
   summary matched "data exfiltration"；has PDF；has rich summary
   原始来源

《Capable but Careless: Do Computer-Use Agents Follow Contextual Integrity?》〔评测 / 应用 / 方法〕：Computer-use agents (CUAs) now act on a user's behalf across personal applications such as email, calendars, and to-do lists. Thi…

1. Capable but Careless: Do Computer-Use Agents Follow Contextual Integrity? · Score 64
   title matched "computer-use agent"；has PDF；has rich summary
   原始来源
2. TROPT: An Open Framework for Unifying and Advancing Discrete Text Optimization · Score 46
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
3. GIF: Locally Sound Geometric Information Flow Control for LLMs · Score 43
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《What Do Safety-Aligned LLMs Learn From Mixed Compliance Demonstrations?》〔方法〕：Prior work has shown that in-context demonstrations can jailbreak language models, but it remains unclear how models interpret different type…

1. What Do Safety-Aligned LLMs Learn From Mixed Compliance Demonstrations? · Score 46
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
2. Analyzing Defensive Misdirection Against Model-Guided Automated Attacks on Agentic AI Systems · Score 46
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
3. RACL: Reasoning-Agent Control Layers for Continuous Metaheuristic Learning · Score 41
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
4. Beyond Static Endpoints: Tool Programs as an Interface for Flexible Agentic Web Services · Score 39
   summary matched "sandboxing"；has PDF；has rich summary
   原始来源

《CodeSentinel: A Three-Layer Defense Against Indirect Prompt Injection in Code Contexts》〔方法〕：Code large language models increasingly retrieve external code context from repositories, documentation, issue threads, and co…

1. CodeSentinel: A Three-Layer Defense Against Indirect Prompt Injection in Code Contexts · Score 108
   title matched "prompt injection"；title matched "indirect prompt injection"；has PDF
   原始来源

《Seeing Is Not Screening: Multimodal Hidden Instruction Attacks on Agent Skill Scanners》〔应用 / 方法〕：Agent skills are emerging as an important attack surface in LLM-based systems. Through an empirical study of existing ski…

1. Seeing Is Not Screening: Multimodal Hidden Instruction Attacks on Agent Skill Scanners · Score 47
   summary matched "privilege escalation"；has PDF；has rich summary
   原始来源
2. A Red-Team Study of Anthropic Fable 5 & Opus 4.8 Models · Score 47
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
3. PreAct: Computer-Using Agents that Get Faster on Repeated Tasks · Score 43
   summary matched "guardrail"；has PDF；has rich summary
   原始来源

《Automated jailbreak attack targeting multiple defense strategies》〔评测 / 方法〕：Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks. However, their safety remains a critical c…

1. Automated jailbreak attack targeting multiple defense strategies · Score 65
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. MyPCBench: A Benchmark for Personally Intelligent Computer-Use Agents · Score 65
   title matched "computer-use agent"；has PDF；has rich summary
   原始来源
3. DoubtProbe: Black-Box Jailbreak Defense via Structural Verification and Semantic Auditing · Score 61
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
4. KVEraser: Learning to Steer KV Cache for Efficient Localized Context Erasing · Score 47
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源
5. Adaptive and Explicit safe: Triggering Latent Safety Awareness in Large Reasoning Models · Score 44
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源

《Neuro-Symbolic Agents for Regulated Process Automation: Challenges and Research Agenda》〔应用 / 方法〕：LLM-based agents are entering regulated industries where they automate judgment intensive quality management processes. W…

1. Neuro-Symbolic Agents for Regulated Process Automation: Challenges and Research Agenda · Score 44
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
2. ComAct: Reframing Professional Software Manipulation via COM-as-Action Paradigm · Score 41
   summary matched "computer-use agent"；has PDF；has rich summary
   原始来源
3. Getting Better at Working With You: Compiling User Corrections into Runtime Enforcement for Coding Agents · Score 40
   summary matched "agent runtime"；has PDF；has rich summary
   原始来源
4. No Hidden Prompts Needed! You Can Game AI Peer Review with Presentation-Only Revisions · Score 38
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源
5. Nous: An Attempt to Extract and Inject the Cognition Behind Prediction-Market Behavior · Score 38
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《Grammar-Constrained Decoding Can Jailbreak LLMs into Generating Malicious Code》〔评测 / 方法〕：Large Language Models (LLMs) are increasingly used for code generation, raising concerns that they may be misused to produce mali…

1. Grammar-Constrained Decoding Can Jailbreak LLMs into Generating Malicious Code · Score 60
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. OCELOT: Inference-Leakage Budgets for Privacy-Preserving LLM Agents · Score 47
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
3. Online Shift Detection and Conformal Adaptation for Deployed Safety Classifiers · Score 41
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
4. External Experience Serving in Production LLM Systems: A Deployment-Oriented Study of Quality-Cost Trade-offs · Score 38
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation》〔评测 / 应用 / 方法〕：Large language model (LLM) agents are rapidly moving from conversational interfaces to software components that plan, invoke t…

1. Toward Secure LLM Agents: Threat Surfaces, Attacks, Defenses, and Evaluation · Score 78
   summary matched "agent security"；summary matched "LLM agent security"；summary matched "prompt injection"
   原始来源
2. Workflow-GYM: Towards Long-Horizon Evaluation of Computer-use Agentic tasks in Real-World Professional Fields · Score 68
   title matched "computer-use agent"；has PDF；has rich summary
   原始来源
3. Data Journalist Agent: Transforming Data into Verifiable Multimodal Stories · Score 48
   summary matched "computer-use agent"；has PDF；has rich summary
   原始来源
4. It Takes One to Bias Them All: Breaking Bad with One-Shot GRPO · Score 45
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
5. Training LLMs to Enforce Multi-Level Instruction Hierarchies via Gravity-Weighted Direct Preference Optimization · Score 44
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces》〔评测 / 方法〕：Computer-use agents (CUAs) increasingly operate in runtimes that combine visual desktop control, command-line ex…

1. WeaveBench: A Long-Horizon, Real-World Benchmark for Computer-Use Agents with Hybrid Interfaces · Score 83
   title matched "computer-use agent"；summary matched "agent runtime"；has PDF
   原始来源
2. Brain-Prompt Injection: A Route-Safety Audit for BCI-LLM Agents · Score 63
   title matched "prompt injection"；has PDF；has rich summary
   原始来源
3. What the Eyes See, the LLMs Miss: Exploiting Human Perception for Adversarial Text Attacks · Score 47
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
4. PRISM: Recovering Instruction Sets from Language Model Activations · Score 45
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《GuardNet: Ensemble Strategies of Shallow Neural Networks for Robust Prompt Injection and Jailbreak Detection》〔评测 / 数据 / 应用 / 方法〕：Large Language Models (LLMs) have transformed natural language processing, but they remai…

1. GuardNet: Ensemble Strategies of Shallow Neural Networks for Robust Prompt Injection and Jailbreak Detection · Score 138
   title matched "prompt injection"；title matched "jailbreak"；summary matched "guardrail"
   原始来源
2. From Risk Classification to Action Plan Remediation: A Guardrail Feedback Driven Framework for LLM Agents · Score 80
   title matched "guardrail"；has PDF；has rich summary
   原始来源
3. Safety Paradox: How Enhanced Safety Awareness Leaves LLMs Vulnerable to Posterior Attack · Score 76
   summary matched "jailbreak"；summary matched "guardrail"；has PDF
   原始来源
4. Beyond Similarity: Trustworthy Memory Search for Personal AI Agents · Score 58
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
5. The Granularity Gap: A Multi-Dimensional Longitudinal Audit of Sycophancy in Gemini Models · Score 58
   summary matched "guardrail"；has PDF；has rich summary
   原始来源

《MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models》〔评测 / 应用 / 方法〕：Diffusion large language models (dLLMs) generate text by iteratively denoising partially masked sequences unde…

1. MaskForge: Structure-Aware Adaptive Attacks for Jailbreaking Diffusion Large Language Models · Score 79
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. What If Prompt Injection Never Left? Exploring Cross-Session Stored Prompt Injection in Agentic Systems · Score 79
   title matched "prompt injection"；has PDF；has rich summary
   原始来源
3. Caught in the Act(ivation): Toward Pre-Output and Multi-Turn Detection of Credential Exfiltration by LLM Agents · Score 75
   summary matched "prompt injection"；summary matched "indirect prompt injection"；has PDF
   原始来源
4. AgentJet: A Flexible Swarm Training Framework for Agentic Reinforcement Learning · Score 57
   summary matched "agent runtime"；has PDF；has rich summary
   原始来源
5. From Untrusted Input to Trusted Memory: A Systematic Study of Memory Poisoning Attacks in LLM Agents · Score 57
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《D-Judge: Disrupting Multi-Turn Jailbreaks using Semantics-Preserving Output Rewriting》〔评测 / 数据 / 方法〕：Multi-turn jailbreak attacks pose a growing threat to large language model (LLM) safety because they exploit feedback…

1. D-Judge: Disrupting Multi-Turn Jailbreaks using Semantics-Preserving Output Rewriting · Score 79
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. MedCUA-Bench: A Screenshot-Only Benchmark for Clinical Computer-Use Agents · Score 79
   title matched "computer-use agent"；has PDF；has rich summary
   原始来源
3. MultiTurnPSB: Evaluating Multi-Turn Jailbreak Attacks an dClassifier-Based Defenses for Medical AI Safety · Score 79
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
4. From Control Boundary to Insurance Claim: Reconstructing AI-Mediated Losses Through the CER Framework · Score 75
   summary matched "prompt injection"；summary matched "malicious tool"；has PDF
   原始来源
5. Acceptance-Test-Driven Evaluation Protocols for Business-Centric LLM Systems · Score 57
   summary matched "guardrail"；has PDF；has rich summary
   原始来源

《Jailbreaking Multimodal Large Language Models using Multi-Clip Video》〔数据 / 应用 / 方法〕：As multimodal large language models (MLLMs) have advanced to process video inputs, concerns have emerged about their potential for mal…

1. Jailbreaking Multimodal Large Language Models using Multi-Clip Video · Score 63
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. SentGuard: Sentence-Level Streaming Guardrails for Large Language Models · Score 62
   title matched "guardrail"；has PDF；has rich summary
   原始来源
3. AgentRedBench: Dynamic Redteaming and Integration-Aware Defense for LLM Agents over SaaS Integrations · Score 61
   summary matched "prompt injection"；summary matched "indirect prompt injection"；has PDF
   原始来源
4. SafeMCP: Proactive Power Regulation for LLM Agent Defense via Environment-Grounded Look-Ahead Reasoning · Score 61
   title matched "agent defense"；has PDF；has rich summary
   原始来源
5. SeClaw: Spec-Driven Security Task Synthesis for Evaluating Autonomous Agents · Score 44
   summary matched "agent security"；has PDF；has rich summary
   原始来源

《Provably Secure Agent Guardrail》〔评测 / 应用 / 方法〕：As large language models transition from bounded generative engines to agents with expansive execution privileges, AI going out of control precipitates a funda…；《Robust an…

1. Provably Secure Agent Guardrail · Score 120
   title matched "secure agent"；title matched "guardrail"；has PDF
   原始来源
2. Robust and Efficient Guardrails with Latent Reasoning · Score 80
   title matched "guardrail"；has PDF；has rich summary
   原始来源
3. AgentDoG 1.5: A Lightweight and Scalable Alignment Framework for AI Agent Safety and Security · Score 58
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
4. Beyond Attack Success Rate: Temporal Logit Observability for LLM Safety Failures · Score 58
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源

《Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents》〔数据 / 方法〕：Computer-use agents (CUAs) have recently made substantial progress, but deploying a separate large expert for each software…

1. Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents · Score 70
   title matched "computer-use agent"；has PDF；has rich summary
   原始来源
2. Code as a Weapon: A Consensus-Labeled Prompt Bank for Measuring Coding-Model Compliance with Malicious-Code Requests · Score 47
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
3. The Ethics of LLM Sandbox and Persona Dynamics · Score 46
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
4. LACUNA: Safe Agents as Recursive Program Holes · Score 46
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源
5. Technical Report: Exploring the Emerging Threats of the Agent Skill Ecosystem · Score 45
   summary matched "data exfiltration"；has PDF；has rich summary
   原始来源

《EviACT: An Evidence-to-Action Framework for Agentic Program Repair》〔评测 / 方法〕：LLM-based agents have moved automated program repair (APR) from fixed-context patch generation to interactive repository-level repair. Howeve…

1. EviACT: An Evidence-to-Action Framework for Agentic Program Repair · Score 122
   summary matched "guardrail"；has PDF；has rich summary
   原始来源
2. Governed Evolution of Agent Runtimes through Executable Operational Cognition · Score 70
   title matched "agent runtime"；has PDF；has rich summary
   原始来源
3. Prompt Injection Detection is Regime-Dependent: A Deployment-Aware Evaluation with Interpretable Structural Signals · Score 65
   title matched "prompt injection"；has PDF；has rich summary
   原始来源
4. BAIT: Boundary-Guided Disclosure Escalation via Self-Conditioned Reasoning · Score 45
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
5. AlbanianLLMSafety: A Safety Evaluation Dataset for Large Language Models in Albanian · Score 43
   summary matched "guardrail"；has PDF；has rich summary
   原始来源

《CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agents》〔评测 / 数据 / 应用 / 方法〕：Reinforcement learning with verifiable rewards (RLVR) has driven breakthroughs in domains such as math, tool-use,…

1. CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agents · Score 62
   title matched "computer-use agent"；has PDF；has rich summary
   原始来源
2. How Agentic AI Coding Assistants Become the Attacker's Shell · Score 44
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源
3. AgentHijack: Benchmarking Computer Use Agent Robustness to Common Environment Corruptions · Score 41
   summary matched "computer-use agent"；has PDF；has rich summary
   原始来源

《DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback》〔评测 / 应用 / 方法〕：LLM-powered AI agents require high-frequency state exploration (e.g., test-time tree search and reinforcement learn…

1. DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback · Score 48
   summary matched "agent sandbox"；has PDF；has rich summary
   原始来源
2. HarnessAPI: A Skill-First Framework for Unified Streaming APIs and MCP Tools · Score 47
   summary matched "agent runtime"；has PDF；has rich summary
   原始来源
3. Contractual Skills: A GovernSpec Design Framework for Enterprise AI Agents · Score 46
   summary matched "guardrail"；has PDF；has rich summary
   原始来源

《Agent JIT Compilation for Latency-Optimizing Web Agent Planning and Scheduling》〔应用 / 方法〕：Computer-use agents (CUA) automate tasks specified with natural language such as "order the cheapest item from Taco Bell" by gene…

1. Agent JIT Compilation for Latency-Optimizing Web Agent Planning and Scheduling · Score 48
   summary matched "computer-use agent"；has PDF；has rich summary
   原始来源

《Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models》〔评测 / 方法〕：Large Reasoning Models (LRMs) have demonstrated remarkable capabilities in solving complex problems by generat…

1. Attention-Guided Reward for Reinforcement Learning-based Jailbreak against Large Reasoning Models · Score 80
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. OpenComputer: Verifiable Software Worlds for Computer-Use Agents · Score 80
   title matched "computer-use agent"；has PDF；has rich summary
   原始来源
3. Robotics-Inspired Guardrails for Foundation Models in Socially Sensitive Domains · Score 80
   title matched "guardrail"；has PDF；has rich summary
   原始来源
4. A Methodology for Selecting and Composing Runtime Architecture Patterns for Production LLM Agents · Score 58
   summary matched "agent runtime"；has PDF；has rich summary
   原始来源
5. Formal Skill: Programmable Runtime Skills for Efficient and Accurate LLM Agents · Score 58
   summary matched "policy enforcement"；has PDF；has rich summary
   原始来源

《An Empirical Study of Privacy Leakage Chains via Prompt Injection in Black-Box Chatbot Environments》〔方法〕：LLM-based chatbot agents increasingly process user requests by combining natural-language reasoning with external…

1. An Empirical Study of Privacy Leakage Chains via Prompt Injection in Black-Box Chatbot Environments · Score 98
   title matched "prompt injection"；summary matched "indirect prompt injection"；summary matched "jailbreak"
   原始来源
2. Multilingual jailbreaking of LLMs using low-resource languages · Score 82
   title matched "jailbreak"；summary matched "guardrail"；has PDF
   原始来源
3. Overeager Coding Agents: Measuring Out-of-Scope Actions on Benign Tasks · Score 68
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源
4. Acoustic Interference: A New Paradigm Weaponizing Acoustic Latent Semantic for Universal Jailbreak against Large Audio Language Models · Score 63
   title matched "jailbreak"；has PDF；has rich summary
   原始来源

《Sleeper Channels and Provenance Gates: Persistent Prompt Injection in Always-on Autonomous AI Agents》〔评测 / 应用 / 方法〕：Always-on AI agents (OpenClaw, Hermes Agent) run as a single persistent process under the owner's iden…

1. Sleeper Channels and Provenance Gates: Persistent Prompt Injection in Always-on Autonomous AI Agents · Score 66
   title matched "prompt injection"；has PDF；has rich summary
   原始来源
2. LLM-Based Persuasion Enables Guardrail Override in Frontier LLMs · Score 63
   title matched "guardrail"；has PDF；has rich summary
   原始来源

《Metaphor Is Not All Attention Needs》〔应用 / 方法〕：Large language models are increasingly deployed in safety-critical applications, where their ability to resist harmful instructions is essential. Although post…；《A microser…

1. Metaphor Is Not All Attention Needs · Score 44
   summary matched "jailbreak"；has PDF；has rich summary
   原始来源
2. A microservices-based endpoint monitoring platform with predictive NLP models for real-time security and hate-speech risk alerting · Score 42
   summary matched "data exfiltration"；has PDF；has rich summary
   原始来源

《Break the Brake, Not the Wheel: Untargeted Jailbreak via Entropy Maximization》〔评测 / 方法〕：Recent studies show that gradient-based universal image jailbreaks on vision-language models (VLMs) exhibit little or no cross-mod…

1. Break the Brake, Not the Wheel: Untargeted Jailbreak via Entropy Maximization · Score 69
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
2. Intrinsic Guardrails: How Semantic Geometry of Personality Interacts with Emergent Misalignment in LLMs · Score 67
   title matched "guardrail"；has PDF；has rich summary
   原始来源
3. Re-Triggering Safeguards within LLMs for Jailbreak Detection · Score 67
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
4. Guaranteed Jailbreaking Defense via Disrupt-and-Rectify Smoothing · Score 67
   title matched "jailbreak"；has PDF；has rich summary
   原始来源
5. RUBEN: Rule-Based Explanations for Retrieval-Augmented LLM Systems · Score 48
   summary matched "prompt injection"；has PDF；has rich summary
   原始来源

《Constraining Host-Level Abuse in Self-Hosted Computer-Use Agents via TEE-Backed Isolation》〔评测 / 应用 / 方法〕：Self-hosted computer-use agents (SHCUAs), such as OpenClaw, combine natural-language interaction with direct acce…

1. Constraining Host-Level Abuse in Self-Hosted Computer-Use Agents via TEE-Backed Isolation · Score 102
   title matched "computer-use agent"；summary matched "prompt injection"；summary matched "indirect prompt injection"
   原始来源