2026年4月发布的多篇研究论文解决了自主AI代理系统中日益增长的安全挑战。这些论文提出了增强交互式AI代理的安全、可信赖和治理的框架和方法论，特别是在网络安全和企业系统等高风险领域。关键主题包括去中心化架构、形式化验证方法、运行时安全执行以及强大的审计机制，以减轻对抗性攻击、数据投毒和未经授权的操作等风险。 AI
影响 这些框架旨在提高AI代理的安全性与可信赖性，可能加速其在关键应用中的采用。
排序理由 多篇在arXiv上发表的研究论文提出了新的人工智能代理安全框架和方法论。
AI 生成摘要 · Google Gemini · 来自 30 个来源。 我们如何撰写摘要 →
arXiv:2605.02682v1 Announce Type: new Abstract: Authorizing Large Language Model (LLM)-driven agents to dynamically invoke tools and access protected resources introduces significant security risks, and the risks grow dramatically as agents engage in multi-turn conversations and …
Authorizing Large Language Model (LLM)-driven agents to dynamically invoke tools and access protected resources introduces significant security risks, and the risks grow dramatically as agents engage in multi-turn conversations and scale toward distributed collaboration. A compro…
arXiv:2604.27464v1 Announce Type: cross Abstract: Autonomous agent frameworks built upon large language models (LLMs) are evolving into complex, tool-integrated, and continuously operating systems, introducing security risks beyond traditional prompt-level vulnerabilities. As thi…
arXiv:2604.26997v1 Announce Type: cross Abstract: Autonomous AI agent ecosystems require stronger mechanisms for secure discovery, identity verification, capability attestation, and policy governance. Current deployments frequently lack (1) uniform agent discovery, (2) cryptograp…
arXiv:2604.27358v1 Announce Type: new Abstract: As large language model (LLM) agents are deployed in high-stakes environments, the question of how safely to delegate subtasks to specialized sub-agents becomes critical. Existing work addresses multi-agent architecture selection at…
arXiv:2602.11897v3 Announce Type: replace-cross Abstract: Cybersecurity decision-making increasingly occurs in environments characterized by uncertainty, partial observability, and adversarial manipulation, where heterogeneous signals from multiple sources are often incomplete, a…
arXiv:2604.27132v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) and Multi-Agent Systems (MAS) in high-stakes domains demand reliable verification, yet centralized approaches suffer four limitations: (1) Robustness, with single points of failure vulnerable to attacks…
arXiv:2505.02077v2 Announce Type: replace-cross Abstract: AI agents are beginning to interact with each other directly and across internet platforms and physical environments, creating security challenges beyond traditional cybersecurity and AI safety frameworks. Free-form protoc…
arXiv:2604.26274v1 Announce Type: cross Abstract: Structured-workflow agents driven by large language models execute tool calls against sensitive external environments. We propose \codename, a telemetry-driven behavioral anomaly detection firewall. Drawing on sequence-based intru…
Structured-workflow agents driven by large language models execute tool calls against sensitive external environments. We propose \codename, a telemetry-driven behavioral anomaly detection firewall. Drawing on sequence-based intrusion detection, \codename\ compiles verified benig…
Enterprise software engineering is shifting away from deterministic CRUD/REST architectures toward AI-native systems where large language models act as cognitive orchestrators. This transition introduces a critical security tension: probabilistic LLMs weaken classical mechanisms …
arXiv:2604.24657v1 Announce Type: cross Abstract: Autonomous AI agents extend large language models into full runtime systems that load skills, ingest external content, maintain memory, plan multi-step actions, and invoke privileged tools. In such systems, security failures rarel…
arXiv:2604.20833v2 Announce Type: replace-cross Abstract: As artificial intelligence (AI) systems are increasingly deployed across critical domains, their security vulnerabilities pose growing risks of high-profile exploits and consequential system failures. Yet systematic approa…
arXiv:2603.09002v2 Announce Type: replace-cross Abstract: Multi-agent artificial intelligence systems or MAS are systems of autonomous agents that exercise delegated tool authority, share persistent memory, and coordinate via inter-agent communication. MAS introduces qualitativel…
arXiv:2604.24020v1 Announce Type: cross Abstract: Autonomous AI agents deployed on platforms such as OpenClaw face prompt injection, memory poisoning, supply-chain attacks, and social engineering, yet existing defences address only the platform perimeter, leaving the agent's own …
arXiv:2604.23459v1 Announce Type: cross Abstract: Multi-agent systems (MAS), composed of networks of two or more autonomous AI agents, have become increasingly popular in production deployments, yet introduce security risks that do not arise in single-agent settings. Even if indi…
arXiv:2604.23338v1 Announce Type: cross Abstract: Agentic AI systems face security challenges that stateless large language models do not. They plan across extended horizons, maintain persistent memory, invoke external tools, and coordinate with peer agents. Existing security ana…
Agent Skills package SKILL.md files, scripts, reference documents, and repository context into reusable capability units, turning pre-load auditing from single-prompt filtering into cross-file security review. Existing guardrails often flag risk but recover malicious intent incon…
Autonomous AI agents extend large language models into full runtime systems that load skills, ingest external content, maintain memory, plan multi-step actions, and invoke privileged tools. In such systems, security failures rarely remain confined to a single interface; instead, …
This report presents a comparative evaluation of DKnownAI Guard in AI agent security scenarios, benchmarked against three competing products: AWS Bedrock Guardrails, Azure Content Safety, and Lakera Guard. Using human annotation as the ground truth, we assess each guardrail's abi…
Autonomous AI agents deployed on platforms such as OpenClaw face prompt injection, memory poisoning, supply-chain attacks, and social engineering, yet existing defences address only the platform perimeter, leaving the agent's own threat judgement entirely untrained. We present Cl…
Recent evidence suggests that frontier AI systems can exhibit agentic misalignment, generating and executing harmful actions derived from internally constructed goals, even without explicit user requests. Existing mitigation methods, such as Reinforcement Learning from Human Feed…
Joint guidance just released from leading Western security agencies on safely adopting agentic AI services. Key considerations, emerging risks, and best practices for implementation. Worth a read. #AI #AgenticAI #Cybersecurity #ASD #ACSC #CISA #NSA #CCCS #NCSCNZ #NCSCUK Careful A…
https://www. europesays.com/2954859/ AI runtime security for the agentic enterprise # AgenticAI # AgenticArtificialIntelligence # AI # ArtificialIntelligence # EnterprisesTurnToRuntimeSecurityToCloseTheAgenticAITrustGap # KellyKnight # SiliconANGLE
https://www. europesays.com/2954857/ SAS cements agentic AI-ready data management foundation # AgenticAI # AgenticArtificialIntelligence # AI # ArtificialIntelligence
CISA, NSA, and partner agencies published fresh guidance on secure AI agent adoption. We analyzed what it means for identity controls, privilege boundaries, and deployment governance in real systems. https:// go.aintelligencehub.com/ma-usc yberagenciesaiagen # AI # Cybersecurity …
Joint guidance just released from leading Western security agencies on safely adopting agentic AI services. Key considerations, emerging risks, and best practices for implementation. Worth a read. #AI #AgenticAI #Cybersecurity #ASD #ACSC #CISA #NSA #CCCS #NCSCNZ #NCSCUK Careful A…
Agentic AI security alert ⚠️ CISA guidance: Autonomous AI systems expand attack surface beyond traditional controls. Risks: • Unchecked actions • API exposure • Data leakage AI = operational entity now. Source: https://www. cisa.gov/resources-tools/resou rces/careful-adoption-age…
Les agents IA autonomes écrivent du code, appellent des APIs, orchestrent des systèmes… souvent sans supervision humaine. La question de confiance dans cette couche "agentique" devient centrale en sécurité. Qui audite ce que l'agent a décidé tout seul à 3h du matin ? 🤖 La surface…
Megent Megent: Open-Source Runtime Policy for AI Agents. Megent helps secure AI agents by enforcing allow/deny tool policies, masking sensitive data, and logging every tool decision. A zero-trust safety layer for AI agents. Wrap tools fast. Ship with confidence. Control every age…