Prompt injection attacks can self-replicate across LLM agents in multi-agent systems, enabling data theft, misinformation, and system disruption while propagating silently.
hub Canonical reference
CAMEL: Communicative Agents for "Mind" Exploration of Large Language Model Society
Canonical reference. 88% of citing Pith papers cite this work as background.
60
Pith papers citing it
99
external citations · Pith
Background
88% of classified citations
abstract
The rapid advancement of chat-based language models has led to remarkable progress in complex task-solving. However, their success heavily relies on human input to guide the conversation, which can be challenging and time-consuming. This paper explores the potential of building scalable techniques to facilitate autonomous cooperation among communicative agents, and provides insight into their "cognitive" processes. To address the challenges of achieving autonomous cooperation, we propose a novel communicative agent framework named role-playing. Our approach involves using inception prompting to guide chat agents toward task completion while maintaining consistency with human intentions. We showcase how role-playing can be used to generate conversational data for studying the behaviors and capabilities of a society of agents, providing a valuable resource for investigating conversational language models. In particular, we conduct comprehensive studies on instruction-following cooperation in multi-agent settings. Our contributions include introducing a novel communicative agent framework, offering a scalable approach for studying the cooperative behaviors and capabilities of multi-agent systems, and open-sourcing our library to support research on communicative agents and beyond: https://github.com/camel-ai/camel.
hub tools
citation-role summary
background 8
citation-polarity summary
roles
background 8representative citing papers
Glite ARF introduces a verifier-driven three-role framework for parallel LLM coding agents, demonstrated by first- and second-place finishes in the BEA 2026 vocabulary-difficulty shared task across three languages with 29.9-35.9% RMSE reduction at ~$450 API cost.
ALMs unify pretrained atomistic encoder, LLM, and denoising diffusion via continuous projectors and staged training to reach SOTA on text-conditioned crystal prediction and de novo generation.
A new 7x4 taxonomy organizes agentic AI security threats by architectural layer and persistence timescale, revealing under-explored upper layers and missing defenses after surveying 116 papers.
SAT trains multi-LLM teams with sequential block updates to deliver monotonic gains and plug-and-play model swaps that provably improve performance bounds.
Discourse among AI agents on Moltbook is largely determined by architectural constraints like context windows and identity files, appearing as social learning but actually short-horizon contextual conditioning.
GAIA benchmark shows humans at 92% accuracy on simple real-world questions far outperform current AI systems at 15%, proposing this gap as a key milestone for general AI.
Current commercial LLMs largely fail multi-agent orchestration prompting: average combined pass rate is about 15%, while GPT-5.5 leads at 62% with still-substantial leakage.
MESA ranks MAS communication edges by vulnerability via graph-theoretic metrics and dynamic probes, achieving mean Spearman ρ=+0.60 correlation with empirical per-edge attack success and 3x interception gain when monitoring the top 10%.
The Novelty-Aware Research Agent layers query analysis, ReAct retrieval, ranking, schema-guided extraction, three-pass comparison, and answer generation on RAG to produce structured comparison artifacts that standard RAG cannot.
AgentMob is a training-free LLM-driven agent that formulates mobility prediction as adaptive evidence-controlled decision making and outperforms other training-free LLM methods on three datasets.
Multi-agent social simulations show LLM privacy violations rising from 19.95% to 45.30%, with leakage spreading contagiously (8x after peer disclosure) and explicit instructions leaving rates above 37.8%.
AgensFlow learns coordination policies from task trajectories and outperforms fixed pipelines on distributed-systems incident and security-advisory tasks.
HierVA improves multi-step chart question answering by having a high-level manager maintain key joint contexts while specialized workers perform targeted reasoning with visual zoom-in.
ASPO combines multi-agent LLM proposals with deterministic enforcement in a MAPE-K loop to select conflict-free, resource-feasible security patterns for IoT, delivering 100% safety invariants and 21-23% tail latency/energy reductions on testbed workloads.
Translating historical governance into LLM multi-agent systems shows institutional topology drives collective performance gaps over 57 points, with optimal forms shifting by model capability and task.
SciCrafter benchmark shows frontier AI agents plateau at 26% success on parameterized Minecraft redstone tasks requiring discovery and application of causal regularities, with knowledge application as the largest gap but gap identification emerging as a new hurdle for top models.
LLMs show poor calibration in predicting task success and token use on software engineering benchmarks, causing market auctions to underperform compared to perfect information scenarios, with limited improvement from added context.
ClawTrace enables cost-aware LLM agent skill distillation by tracing per-step costs and generating preserve, prune, and repair patches, with ablations showing reduced regressions and prune rules transferring to cut costs by 32%.
PBRC is a contract protocol that enforces evidential belief updates in deliberative multi-agent systems and proves it prevents conformity-driven false cascades under conservative fallbacks.
AgentCity introduces a Separation of Power constitutional architecture on blockchain for governing autonomous agent economies through agent legislation, automated execution, and human accountability.
Multi-LLM committees amplify small input perturbations into divergent deliberation trajectories and decisions under deterministic conditions.
A device-native autonomous agent system using zero-knowledge proofs and distilled world models achieves 87% negotiation success, 2.4x lower latency than cloud systems, and 27% higher user trust in privacy-sensitive scenarios.
WM-SAR decomposes sarcasm into LLM-agent components, quantifies literal-normative inconsistency deterministically, and integrates it with intention via logistic regression to outperform prior sarcasm detectors on benchmarks.
citing papers explorer
-
Prompt Infection: LLM-to-LLM Prompt Injection within Multi-Agent Systems
Prompt injection attacks can self-replicate across LLM agents in multi-agent systems, enabling data theft, misinformation, and system disruption while propagating silently.
-
Glite ARF: Verifier-Driven Research with Parallel LLM Coding Agents
Glite ARF introduces a verifier-driven three-role framework for parallel LLM coding agents, demonstrated by first- and second-place finishes in the BEA 2026 vocabulary-difficulty shared task across three languages with 29.9-35.9% RMSE reduction at ~$450 API cost.
-
Atomistic Language Models Understand and Generate Materials
ALMs unify pretrained atomistic encoder, LLM, and denoising diffusion via continuous projectors and staged training to reach SOTA on text-conditioned crystal prediction and de novo generation.
-
A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework
A new 7x4 taxonomy organizes agentic AI security threats by architectural layer and persistence timescale, revealing under-explored upper layers and missing defenses after surveying 116 papers.
-
SAT: Sequential Agent Tuning for Coordinator Free Plug and Play Multi-LLM Training with Monotonic Improvement Guarantees
SAT trains multi-LLM teams with sequential block updates to deliver monotonic gains and plug-and-play model swaps that provably improve performance bounds.
-
What Do AI Agents Talk About? Discourse and Architectural Constraints in the First AI-Only Social Network
Discourse among AI agents on Moltbook is largely determined by architectural constraints like context windows and identity files, appearing as social learning but actually short-horizon contextual conditioning.
-
GAIA: a benchmark for General AI Assistants
GAIA benchmark shows humans at 92% accuracy on simple real-world questions far outperform current AI systems at 15%, proposing this gap as a key milestone for general AI.
-
PerspectiveGap: A Benchmark for Multi-Agent Orchestration Prompting
Current commercial LLMs largely fail multi-agent orchestration prompting: average combined pass rate is about 15%, while GPT-5.5 leads at 62% with still-substantial leakage.
-
MESA: Prioritizing Vulnerable Communication Channels for Securing Multi-Agent Systems
MESA ranks MAS communication edges by vulnerability via graph-theoretic metrics and dynamic probes, achieving mean Spearman ρ=+0.60 correlation with empirical per-edge attack success and 3x interception gain when monitoring the top 10%.
-
Novelty-Aware Agentic Retrieval: Comparing Research Contributions Through Structured Multi-Step Reasoning
The Novelty-Aware Research Agent layers query analysis, ReAct retrieval, ranking, schema-guided extraction, three-pass comparison, and answer generation on RAG to produce structured comparison artifacts that standard RAG cannot.
-
Towards Efficient and Evidence-grounded Mobility Prediction with LLM-Driven Agent
AgentMob is a training-free LLM-driven agent that formulates mobility prediction as adaptive evidence-controlled decision making and outperforms other training-free LLM methods on three datasets.
-
Got a Secret? LLM Agents Can't Keep It: Evaluating Privacy in Multi-Agent Systems
Multi-agent social simulations show LLM privacy violations rising from 19.95% to 45.30%, with leakage spreading contagiously (8x after peer disclosure) and explicit instructions leaving rates above 37.8%.
-
AgensFlow: A Coordination-Policy Substrate for Multi-Agent Systems
AgensFlow learns coordination policies from task trajectories and outperforms fixed pipelines on distributed-systems incident and security-advisory tasks.
-
Hierarchical Visual Agent: Managing Contexts in Joint Image-Text Space for Advanced Chart Reasoning
HierVA improves multi-step chart question answering by having a high-level manager maintain key joint contexts while specialized workers perform targeted reasoning with visual zoom-in.
-
Self-Adaptive Multi-Agent LLM-Based Security Pattern Selection for IoT Systems
ASPO combines multi-agent LLM proposals with deterministic enforcement in a MAPE-K loop to select conflict-free, resource-feasible security patterns for IoT, delivering 100% safety invariants and 21-23% tail latency/energy reductions on testbed workloads.
-
When Agents Evolve, Institutions Follow
Translating historical governance into LLM multi-agent systems shows institutional topology drives collective performance gaps over 57 points, with optimal forms shifting by model capability and task.
-
Can Current Agents Close the Discovery-to-Application Gap? A Case Study in Minecraft
SciCrafter benchmark shows frontier AI agents plateau at 26% success on parameterized Minecraft redstone tasks requiring discovery and application of causal regularities, with knowledge application as the largest gap but gap identification emerging as a new hurdle for top models.
-
MarketBench: Evaluating AI Agents as Market Participants
LLMs show poor calibration in predicting task success and token use on software engineering benchmarks, causing market auctions to underperform compared to perfect information scenarios, with limited improvement from added context.
-
ClawTrace: Cost-Aware Tracing for LLM Agent Skill Distillation
ClawTrace enables cost-aware LLM agent skill distillation by tracing per-step costs and generating preserve, prune, and repair patches, with ablations showing reduced regressions and prune rules transferring to cut costs by 32%.
-
Preregistered Belief Revision Contracts
PBRC is a contract protocol that enforces evidential belief updates in deliberative multi-agent systems and proves it prevents conformity-driven false cascades under conservative fallbacks.
-
AgentCity: Constitutional Governance for Autonomous Agent Economies via Separation of Power
AgentCity introduces a Separation of Power constitutional architecture on blockchain for governing autonomous agent economies through agent legislation, automated execution, and human accountability.
-
Collective AI can amplify tiny perturbations into divergent decisions
Multi-LLM committees amplify small input perturbations into divergent deliberation trajectories and decisions under deterministic conditions.
-
Device-Native Autonomous Agents for Privacy-Preserving Negotiations
A device-native autonomous agent system using zero-knowledge proofs and distilled world models achieves 87% negotiation success, 2.4x lower latency than cloud systems, and 27% higher user trust in privacy-sensitive scenarios.
-
World model inspired sarcasm reasoning with large language model agents
WM-SAR decomposes sarcasm into LLM-agent components, quantifies literal-normative inconsistency deterministically, and integrates it with intention via logistic regression to outperform prior sarcasm detectors on benchmarks.
-
Dynamic Generation of Multi-LLM Agents Communication Topologies with Graph Diffusion Models
GTD generates task-adaptive, sparse communication topologies for multi-LLM agents via guided iterative graph diffusion steered by a proxy model predicting accuracy, utility, and cost.
-
MAmmoTH: Building Math Generalist Models through Hybrid Instruction Tuning
MAmmoTH models trained via hybrid CoT-PoT instruction tuning on MathInstruct outperform prior open-source LLMs by 16-32% average accuracy on nine math datasets, reaching 33% and 44% on MATH for 7B and 34B scales.
-
A Survey on Large Language Model based Autonomous Agents
A survey of LLM-based autonomous agents that proposes a unified framework for their construction and reviews applications in social science, natural science, and engineering along with evaluation methods and future directions.
-
ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate
Multi-agent debate among LLMs yields more reliable text evaluations than single-agent prompting by simulating collaborative human judgment.
-
Enhancing Chat Language Models by Scaling High-quality Instructional Conversations
UltraChat supplies 1.5 million high-quality multi-turn dialogues that, when used to fine-tune LLaMA, produce UltraLLaMA, which outperforms prior open-source chat models including Vicuna.
-
PaperClaw: Harnessing Agents for Autonomous Research and Human-in-the-Loop Refinement
PAPERCLAW is a multi-agent system for end-to-end autonomous research paper generation from literature to output, with human refinement and LLM-judge evaluation showing strong results.
-
OpenRath: Session-Centered Runtime State for Agent Systems
OpenRath introduces Session as a first-class, branchable runtime value that unifies fragmented state in multi-agent systems and makes fork, merge, and replay explicit operations.
-
A Technical Taxonomy of LLM Agent Communication Protocols
Creates a five-dimension taxonomy (counterparty, payload, interaction state, discovery mechanism, schema flexibility) from nine protocols and identifies architectural patterns plus convergence trends.
-
The Internet of Agentic AI: Communication, Coordination, and Collective Intelligence at Scale
Outlines a vision and key research challenges for scalable networks of autonomous AI agents drawing on multi-agent systems, networks, and security.
-
ConMem: Structured Memory-Guided Adaptation in Training-Free Multi-Agent Systems
ConMem distills agent trajectories into structured memory cards organized in a relation-aware graph to enable training-free, relation-coordinated adaptation in LLM-based multi-agent systems.
-
Beyond tokens: a unified framework for latent communication in LLM-based multi-agent systems
A taxonomy that places 18 latent-communication methods for LLM agents on WHAT/WHICH/HOW axes and extracts design trade-offs.
-
Do More Agents Help? Controlled and Protocol-Aligned Evaluation of LLM Agent Workflows
Under controlled identical protocols, only one of six multi-agent LLM systems marginally exceeds a single-agent baseline on benchmark-balanced accuracy while the rest trail and cost more; a runtime workflow reaches 66.72% on GAIA.
-
What Should Agents Say? Action-state Communication for Efficient Multi-Agent Systems
Introduces PACT protocol that projects agent outputs into action-state records, yielding comparable or better task performance with substantially fewer tokens in multi-agent LLM systems and production harnesses.
-
LipoAgent: Coordinating Fine-Tuned LLM Agents for Safer Lipid Design
LipoAgent coordinates fine-tuned LLM agents with a toxicity-first conditional prediction to improve lipid design for safer mRNA delivery, reporting 32% relative gains and wet-lab confirmation of virtual rankings.
-
Do LLM Agents Mirror Socio-Cognitive Effects in Power-Asymmetric Conversations?
LLMs assigned high or low status personas in multi-turn dialogues exhibit socio-cognitive effects including language coordination, pronoun patterns, persuasion success, and compliance with unsafe requests.
-
Intermediate Artifacts as First-Class Citizens: A Data Model for Durable Intermediate Artifacts in Agentic Systems
A systems-level data model for preserving typed, addressable, versioned, and dependency-aware intermediate artifacts in agentic AI systems to improve long-term inspectability and maintainability.
-
Beyond Inefficiency: Systemic Costs of Incivility in Multi-Agent Monte Carlo Simulations
Monte Carlo simulations of LLM agents confirm that toxic debates take 25% longer to converge, with larger delays in smaller models, and show a first-mover advantage independent of toxicity.
-
From Agent Loops to Deterministic Graphs: Execution Lineage for Reproducible AI-Native Work
Execution lineage models AI-native work as a DAG of computations with explicit dependencies, achieving perfect state preservation in controlled update tasks where loop-based agents introduce churn and contamination.
-
Frame Entrepreneurs in an AI Agent Community: Concentrated Identity-Claim Production on Moltbook
In the Moltbook AI agent community, identity-claim production is highly concentrated among a few frame entrepreneurs, with event-driven attention not translating into broad claim-making.
-
Evaluating LLM Coding Agents on SZ-Family Lossy Compression Across Architectures
LLM coding agents show cross-architecture divergence: higher GPU throughput with Gemini but prompt-sensitivity, while GPT is steadier; on Cerebras, producing runnable code dominates over performance tuning, with GPT succeeding where Gemini fails.
-
Multi-Drafter Speculative Decoding with Alignment Feedback
MetaSD integrates multiple heterogeneous drafters into speculative decoding, dynamically selecting them via alignment feedback modeled as a multi-armed bandit to consistently outperform single-drafter baselines.
-
NetworkGames: Simulating Cooperation in Network Games with Personality-driven LLM Agents
Simulations show that cooperative outcomes in network games with personality-driven LLM agents depend on both network connectivity and the placement of pro-social personalities, not just pairwise interaction preferences.
-
What makes a harness a harness: necessary and sufficient conditions for an agent harness
Proposes and tests a constitutive definition of 'agent harness' via conceptual analysis of literature and six real systems.
-
Queen-Bee Agents: A BeeSpec-Centered Architecture for Governed Enterprise MCP Orchestration
Queen-Bee architecture uses a Queen control plane and BeeSpec for governed MCP orchestration, with prototype showing 0.964 success rate and zero governance failures on 59 tasks.
-
Towards Cybersecurity SuperIntelligence (CSI): What's the best harness for cybersecurity?
CSI meta-scaffold unifies five LLM agent harnesses; a blackboard multi-agent system solves 19/33 cybench challenges (57.6%) versus 15/33 for the best single scaffold.
-
Context: Proactive Goal-Directed Intelligence via Composable Sandboxed Programs, Declarative Wiring, and Structured Interaction
An architecture for proactive goal-directed AI agents is presented with six formal theorems claiming Pareto improvements over reactive chatbots in multi-participant task settings.