볼트 홈: README · 마스터 서지: sources

전체 논문 인덱스 — arXiv 96편

이 볼트가 인용한 arXiv 논문 96편을 챕터별로 모았다. ID를 누르면 arxiv.org 원문으로, “인용 노트”의 위키링크를 누르면 그 논문을 다루는 볼트 노트로 넘어간다. 이 가운데 22편은 2026-06-20 보강에서 추가한 2026년 신작이다.

챕터 배정 기준은 2026년 신작 22편은 주제, 나머지는 인용 노트다. 한 논문이 여러 챕터에서 인용되면 대표 챕터에만 싣고, 인용 노트 열에는 관련 노트를 모두 표기했다.

00 개요 (32편)

arXiv제목인용 노트
1706.03762Attention Is All You Need00_03-Transformer-어텐션-KV캐시
2209.11895In-context Learning and Induction Heads
2212.08073Constitutional AI: Harmlessness from AI Feedback
2305.14325Improving Factuality and Reasoning in Language Models through Multiagent Debate
2311.09198Never Lost in the Middle: Mastering Long-Context Question Answering with Position-Agnostic Decompositional Training
2403.04797Found in the Middle: How Language Models Use Long Contexts Better via Plug-and-Play Positional Encoding
2410.02694HELMET: How to Evaluate Long-Context Language Models Effectively and Thoroughly
2410.10813LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory
2411.03538Long Context RAG Performance of Large Language Models00_04-컨텍스트-로트-Context-Rot · 06_10-RAG-vs-롱컨텍스트-논쟁
2411.15399Less is More: Optimizing Function Calling for LLM Execution on Edge Devices00_06-5대-실패모드-분류 · 00_실패모드-진단-퀴즈 · 03_03-소형모델-취약성
2412.19442A Survey on Large Language Model Acceleration based on KV Cache Management00_03-Transformer-어텐션-KV캐시
2501.06322Multi-Agent Collaboration Mechanisms: A Survey of LLMs
2502.08177SycEval: Evaluating LLM Sycophancy
2502.19130Voting or Consensus? Decision-Making in Multi-Agent Debate
2504.00180Contradiction Detection in RAG Systems: Evaluating LLMs as Context Validators for Improved Information Consistency
2505.02077Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agents
2505.02279A survey of agent interoperability protocols: Model Context Protocol (MCP), Agent Communication Protocol (ACP), Agent-to-Agent Protocol (A2A), and Agent Network Protocol (ANP)
2505.06120LLMs Get Lost In Multi-Turn Conversation00_06-5대-실패모드-분류 · 00_실패모드-진단-퀴즈 · 00_핵심용어집
2505.15392Understanding the Anchoring Effect of LLM with Synthetic Data: Existence, Mechanism, and Potential Mitigations
2507.06261Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities00_06-5대-실패모드-분류 · 00_실패모드-진단-퀴즈 · 04_09-에이전트-취약성-왜-더-심각한가
2510.00615ACON: Optimizing Context Compression for Long-horizon LLM Agents
2510.26493Context Engineering 2.0: The Context of Context Engineering00_01-용어-탄생과-계보
2601.02371Permission Manifests for Web Agents
2602.22724AgentSentry: Mitigating Indirect Prompt Injection in LLM Agents via Temporal Causal Diagnostics and Context Purification
2603.03308Old Habits Die Hard: How Conversational History Geometrically Traps LLMs
2603.10123Lost in the Middle at Birth: An Exact Theory of Transformer Position Bias
2603.20397KV Cache Optimization Strategies for Scalable and Efficient LLM Inference00_03-Transformer-어텐션-KV캐시
2604.01664ContextBudget: Budget-Aware Context Management for Long-Horizon Search Agents00_01-용어-탄생과-계보 · 00_04-컨텍스트-로트-Context-Rot · 00_05-에이전트-컨텍스트-누적 · 00_07-4대-전략-Write-Select-Compress-Isolate
2604.07007AgentCity: Constitutional Governance for Autonomous Agent Economies via Separation of Power
2604.07911Dynamic Attentional Context Scoping: Agent-Triggered Focus Sessions for Isolated Per-Agent Steering in Multi-Agent LLM Orchestration
2604.16339Semantic Consensus: Process-Aware Conflict Detection and Resolution for Enterprise Multi-Agent LLM Systems
2606.20245Navigating Unreliable Parametric and Contextual Knowledge: Explicit Knowledge Conflict Resolution for LLM Inference

01 오염 (16편)

arXiv제목인용 노트
2402.07867PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models01_04-우발-vs-적대-비교 · 01_07-paper-memorygraft · 01_08-적대적-오염-위협지형 · 01_10-논쟁점-미해결과제
2407.12784AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases01_04-우발-vs-적대-비교 · 01_07-paper-memorygraft · 01_08-적대적-오염-위협지형 · 01_10-논쟁점-미해결과제
2503.18813Defeating Prompt Injections by Design01_00_MOC · 01_04-우발-vs-적대-비교 · 01_05-paper-agentpoison · 01_07-paper-memorygraft · 01_08-적대적-오염-위협지형 · 01_09-처방-방어전략 · 01_11-paper-CaMeL-방어 · 07_08-처방-컨텍스트는-슬롯계약-4계층-모델
2506.23260From Prompt Injections to Protocol Exploits: Threats in LLM-Powered AI Agents Workflows01_05-paper-agentpoison · 01_08-적대적-오염-위협지형
2509.10540EchoLeak: The First Real-World Zero-Click Prompt Injection Exploit in a Production LLM System01_12-사례-산업-실사고
2512.16962MemoryGraft: Persistent Compromise of LLM Agents via Poisoned Experience Retrieval01_04-우발-vs-적대-비교 · 01_07-paper-memorygraft
2601.05504Memory Poisoning Attack and Defense on Memory Based LLM-Agents01_09-처방-방어전략 · 01_10-논쟁점-미해결과제
2601.05755VIGIL: Defending LLM Agents Against Tool Stream Injection via Verify-Before-Commit09_09-해설-오염-깊이읽기
2601.10923Hidden-in-Plain-Text: A Benchmark for Social-Web Indirect Prompt Injection in RAG09_09-해설-오염-깊이읽기
2602.04711Addressing Corpus Knowledge Poisoning Attacks on RAG Using Sparse Attention09_09-해설-오염-깊이읽기
2602.15654Zombie Agents: Persistent Control of Self-Evolving LLM Agents via Self-Reinforcing Injections09_09-해설-오염-깊이읽기
2602.21447Adversarial Intent is a Latent Variable: Stateful Trust Inference for Securing Multimodal Agentic RAG09_09-해설-오염-깊이읽기
2603.11768Governing Evolving Memory in LLM Agents: Risks, Mechanisms, and the Stability and Safety Governed Memory (SSGM) Framework01_02-메커니즘-자기강화-루프 · 01_04-우발-vs-적대-비교 · 01_09-처방-방어전략 · 01_10-논쟁점-미해결과제
2604.07536TRUSTDESC: Preventing Tool Poisoning in LLM Applications via Trusted Description Generation09_09-해설-오염-깊이읽기
2604.08304Securing Retrieval-Augmented Generation: A Taxonomy of Attacks, Defenses, and Future Directions09_09-해설-오염-깊이읽기
2606.04329From Untrusted Input to Trusted Memory: A Systematic Study of Memory Poisoning Attacks in LLM Agents09_09-해설-오염-깊이읽기

02 산만 (5편)

03 혼란 (9편)

arXiv제목인용 노트
2505.03275RAG-MCP: Mitigating Prompt Bloat in LLM Tool Selection via Retrieval-Augmented Generation03_05-paper-rag-mcp
2505.06416ScaleMCP: Dynamic and Auto-Synchronizing Model Context Protocol Tools for LLM Agents03_05-paper-rag-mcp
2512.24565MCPAgentBench: A Real-world Task Benchmark for Evaluating LLM Agent MCP Tool Use09_11-해설-혼란-깊이읽기
2601.05214Internal Representations as Indicators of Hallucinations in Agent Tool Selection03_08-해결전략-less-is-more
2602.00933MCP-Atlas: A Large-Scale Benchmark for Tool-Use Competency with Real MCP Servers09_11-해설-혼란-깊이읽기
2602.20426Learning to Rewrite Tool Descriptions for Reliable LLM-Agent Tool Use03_08-해결전략-less-is-more
2605.24660How Many Tools Should an LLM Agent See? A Chance-Corrected Answer03_03-소형모델-취약성 · 03_06-paper-cmtf-bor
2606.06284ToolChoiceConfusion: Causal Minimal Tool Filtering for Reliable LLM Agents03_06-paper-cmtf-bor
2606.10209Less Context, Better Agents: Efficient Context Engineering for Long-Horizon Tool-Using LLM Agents03_07-paper-less-context-better-agents

04 충돌 (10편)

arXiv제목인용 노트
2511.04694Reasoning Up the Instruction Ladder for Controllable Language Models04_09-에이전트-취약성-왜-더-심각한가
2602.04288Contextual Drag: How Errors in the Context Affect LLM Reasoning04_09-에이전트-취약성-왜-더-심각한가
2603.00024Personalization Increases Affective Alignment but Has Role-Dependent Effects on Epistemic Independence in LLMs09_12-해설-충돌-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트
2603.10521IH-Challenge: A Training Dataset to Improve Instruction Hierarchy on Frontier LLMs09_12-해설-충돌-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트
2603.12123Cross-Context Review: Improving LLM Output Quality by Separating Production and Review Sessions04_09-에이전트-취약성-왜-더-심각한가
2603.16152HIPO: Instruction Hierarchy via Constrained Reinforcement Learning09_12-해설-충돌-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트
2604.09075Hierarchical Alignment: Enforcing Hierarchical Instruction-Following in LLMs through Logical Consistency09_12-해설-충돌-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트
2604.09443Many-Tier Instruction Hierarchy in LLM Agents04_09-에이전트-취약성-왜-더-심각한가
2605.27784Diagnosing Live Within-Policy Instruction Conflicts in LLM Agents with Witnessed Resolution Profiles09_12-해설-충돌-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트
2606.10860Training LLMs to Enforce Multi-Level Instruction Hierarchies via Gravity-Weighted Direct Preference Optimization09_12-해설-충돌-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트

05 부패 (9편)

arXiv제목인용 노트
2104.09864RoFormer: Enhanced Transformer with Rotary Position Embedding05_03-메커니즘-왜-발생하는가
2302.00093Large Language Models Can Be Easily Distracted by Irrelevant Context05_01-정의-컨텍스트-부패 · 05_02-분류-다른-실패모드와의-관계 · 05_04-paper-context-rot-chroma · 05_07-처방-완화-체크리스트
2307.03172Lost in the Middle: How Language Models Use Long Contexts05_02-분류-다른-실패모드와의-관계 · 05_03-메커니즘-왜-발생하는가 · 05_07-처방-완화-체크리스트
2404.06654RULER: What’s the Real Context Size of Your Long-Context Language Models?02_10-논쟁-산만-vs-훈련부족-vs-아키텍처 · 05_01-정의-컨텍스트-부패 · 05_02-분류-다른-실패모드와의-관계 · 05_04-paper-context-rot-chroma
2502.05167NoLiMa: Long-Context Evaluation Beyond Literal Matching02_10-논쟁-산만-vs-훈련부족-vs-아키텍처 · 05_01-정의-컨텍스트-부패 · 05_02-분류-다른-실패모드와의-관계 · 05_03-메커니즘-왜-발생하는가 · 05_04-paper-context-rot-chroma
2506.11440AbsenceBench: Language Models Can’t Tell What’s Missing05_01-정의-컨텍스트-부패 · 05_03-메커니즘-왜-발생하는가
2601.11564Context Discipline and Performance Correlation: Analyzing LLM Performance and Quality Degradation Under Varying Context Lengths09_10-해설-산만-깊이읽기 · 09_13-해설-부패-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트
2601.15300Intelligence Degradation in Long-Context LLMs: Critical Threshold Determination via Natural Length Distribution Analysis09_13-해설-부패-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트
2605.12366Classifier Context Rot: Monitor Performance Degrades with Context Length09_13-해설-부패-깊이읽기 · 09_14-2026-최신동향-5대모드-업데이트

06 해결전략 (10편)

arXiv제목인용 노트
2304.08467Learning to Compress Prompts with Gist Tokens06_03-Compress전략-요약-프루닝-KV캐시최적화
2307.13854WebArena: A Realistic Web Environment for Building Autonomous Agents06_11-에이전트-벤치마크-tau-swe-gaia
2310.05736LLMLingua: Compressing Prompts for Accelerated Inference of Large Language Models06_03-Compress전략-요약-프루닝-KV캐시최적화
2310.06770SWE-bench: Can Language Models Resolve Real-World GitHub Issues?06_11-에이전트-벤치마크-tau-swe-gaia
2310.08560MemGPT: Towards LLMs as Operating Systems06_01-Write전략-스크래치패드-파일시스템-크로스세션 · 06_08-MemGPT-Letta-OS형-롱텀메모리 · 06_09-실패모드-전략-매핑표
2311.12983GAIA: a benchmark for General AI Assistants06_11-에이전트-벤치마크-tau-swe-gaia
2406.12045$τ$-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains06_11-에이전트-벤치마크-tau-swe-gaia
2407.16833Retrieval Augmented Generation or Long-Context LLMs? A Comprehensive Study and Hybrid Approach06_10-RAG-vs-롱컨텍스트-논쟁
2506.07982$τ^2$-Bench: Evaluating Conversational Agents in a Dual-Control Environment06_11-에이전트-벤치마크-tau-swe-gaia
2507.13334A Survey of Context Engineering for Large Language Models06_09-실패모드-전략-매핑표

07 거버넌스 (5편)

arXiv제목인용 노트
2602.17913From Lossy to Verified: A Provenance-Aware Tiered Memory for Agents07_11-provenance-운영표준-C2PA-VC
2603.07670Memory for Autonomous LLM Agents:Mechanisms, Evaluation, and Emerging Frontiers09_14-2026-최신동향-5대모드-업데이트
2603.09619Context Engineering: From Prompts to Corporate Multi-Agent Architecture09_14-2026-최신동향-5대모드-업데이트
2603.18043The Provenance Paradox in Multi-Agent LLM Routing: Delegation Contracts and Attested Identity in LDP07_06-provenance와-신뢰등급-프로베넌스-역설
2604.16548A Survey on Long-Term Memory Security in LLM Agents: Attacks, Defenses, and Governance Across the Memory Lifecycle09_14-2026-최신동향-5대모드-업데이트

관련