惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

爱范儿
爱范儿
大猫的无限游戏
大猫的无限游戏
WordPress大学
WordPress大学
C
Cyber Attacks, Cyber Crime and Cyber Security
D
DataBreaches.Net
G
Google Developers Blog
博客园 - Franky
V
V2EX
博客园 - 叶小钗
D
Docker
The GitHub Blog
The GitHub Blog
Microsoft Security Blog
Microsoft Security Blog
博客园 - 【当耐特】
H
Hackread – Cybersecurity News, Data Breaches, AI and More
B
Blog RSS Feed
月光博客
月光博客
M
MIT News - Artificial intelligence
F
Fortinet All Blogs
Microsoft Azure Blog
Microsoft Azure Blog
人人都是产品经理
人人都是产品经理
IT之家
IT之家
Google DeepMind News
Google DeepMind News
Apple Machine Learning Research
Apple Machine Learning Research
V
Visual Studio Blog
博客园 - 司徒正美
Stack Overflow Blog
Stack Overflow Blog
罗磊的独立博客
J
Java Code Geeks
U
Unit 42
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
博客园 - 聂微东
T
Tailwind CSS Blog
T
The Blog of Author Tim Ferriss
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
Blog — PlanetScale
Blog — PlanetScale
Jina AI
Jina AI
C
Check Point Blog
Y
Y Combinator Blog
MyScale Blog
MyScale Blog
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
阮一峰的网络日志
阮一峰的网络日志
宝玉的分享
宝玉的分享
B
Blog
小众软件
小众软件
云风的 BLOG
云风的 BLOG
I
InfoQ
Recorded Future
Recorded Future
酷 壳 – CoolShell
酷 壳 – CoolShell
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
美团技术团队

cs.CR updates on arXiv.org

Backdoor Channels Hidden in Latent Space: Cryptographic Undetectability in Modern Neural Networks CTFusion: A CTF-based Benchmark for LLM Agent Evaluation Large Language Models for Agentic NetOps and AIOps: Architectures, Evaluation, and Safety From Controlled to the Wild: Evaluation of Pentesting Agents for the Real-World Red-Teaming Agent Execution Contexts: Open-World Security Evaluation on OpenClaw Graph Representation Learning Augmented Model Manipulation on Federated Fine-Tuning of LLMs Containment Verification: AI Safety Guarantees Independent of Alignment Defense effectiveness across architectural layers: a mechanistic evaluation of persistent memory attacks on stateful LLM agents From Specification to Deployment: Empirical Evidence from a W3C VC + DID Trust Infrastructure for Autonomous Agents Agentic Vulnerability Reasoning on Windows COM Binaries From Beats to Breaches:How Offensive AI Infers Sensitive User Information from Playlists Undetectable Backdoors in Model Parameters: Hiding Sparse Secrets in High Dimensions When Embedding-Based Defenses Fail: Rethinking Safety in LLM-Based Multi-Agent Systems When Alignment Isn't Enough: Response-Path Attacks on LLM Agents Block-wise Codeword Embedding for Reliable Multi-bit Text Watermarking FlexServe: A Fast and Secure LLM Serving System for Mobile Devices with Flexible Resource Isolation TwoHamsters: Benchmarking Multi-Concept Compositional Unsafety in Text-to-Image Models Symbolic Guardrails for Domain-Specific Agents: Stronger Safety and Security Guarantees Without Sacrificing Utility Hardening x402: PII-Safe Agentic Payments via Pre-Execution Metadata Filtering Hijacking Text Heritage: Hiding the Human Signature through Homoglyphic Substitution Like a Hammer, It Can Build, It Can Break: Large Language Model Uses, Perceptions, and Adoption in Cybersecurity Operations on Reddit Private Seeds, Public LLMs: Realistic and Privacy-Preserving Synthetic Data Generation Chimera: Neuro-Symbolic Attention Primitives for Trustworthy Dataplane Intelligence Sockpuppetting: Jailbreaking LLMs by Combining Prefilling with Optimization StegoStylo: Squelching Stylometric Scrutiny through Steganographic Stitching Learning-Based Automated Adversarial Red-Teaming for Robustness Evaluation of Large Language Models AutoGraphAD: Unsupervised network anomaly detection using Variational Graph Autoencoders CrossGuard: Safeguarding MLLMs against Joint-Modal Implicit Malicious Attacks Feedback Lunch: Learned Feedback Codes for Secure Communications Noise Aggregation Analysis Driven by Small-Noise Injection: Efficient Membership Inference for Diffusion Models A First Look at the Security Issues in the Model Context Protocol Ecosystem Formalizing the Safety, Security, and Functional Properties of Agentic AI Systems MEASER: Malware embedding attacks on open-source LLMs Fall into a Pit, Gain in a Wit: Cognitive-Guided Harmful Meme Detection via Misjudgment Risk Pattern Retrieval When Search Goes Wrong: Red-Teaming Web-Augmented Large Language Models Differentially Private Synthetic Text Generation for Retrieval-Augmented Generation (RAG) From surveillance to signalling: escalation channels as environmental controls for agentic AI STAC: When Innocent Tools Form Dangerous Chains to Jailbreak LLM Agents Federated Spatiotemporal Graph Learning for Passive Attack Detection in Smart Grids Guidance Watermarking for Diffusion Models SecureVibeBench: Benchmarking Secure Vibe Coding of AI Agents via Reconstructing Vulnerability-Introducing Scenarios xOffense: An Autonomous Multi-Agent Framework for Penetration Testing with Domain-Adapted Large Language Models Hammer and Anvil: Toward a Theory of Backdoors in Federated Learning Neuro-Symbolic AI for Cybersecurity: State of the Art, Challenges, and Opportunities Tell-Tale Watermarks for Explanatory Reasoning in Synthetic Media Forensics Between a Rock and a Hard Place: The Tension Between Ethical Reasoning and Safety Alignment in LLMs A Comprehensive Guide to Differential Privacy: From Theory to User Expectations Enabling Transparent Cyber Threat Intelligence Combining Large Language Models and Domain Ontologies Unveiling Unicode's Unseen Underpinnings in Undermining Authorship Attribution Searching for Privacy Risks in LLM Agents via Simulation SPRINT: Robust Model Attribution of Generated Images via Secret Pixel Reconstruction Majority Bit-Aware Watermarking For Large Language Models Coward: Collision-based OOD Watermarking for Practical Proactive Federated Backdoor Detection Prompt to Pwn: Automated Exploit Generation for Smart Contracts Activation-Guided Local Editing for Jailbreaking Attacks Random Walk Learning and the Pac-Man Attack ExCyTIn-Bench: Evaluating LLM agents on Cyber Threat Investigation White-Basilisk: A Hybrid Model for Code Vulnerability Detection Intrinsic Fingerprint of LLMs: Continue Training is NOT All You Need to Steal A Model! InvisibleInk: High-Utility and Low-Cost Text Generation with Differential Privacy Logit-Gap Steering: A Forward-Pass Diagnostic for Alignment Robustness Toward Principled LLM Safety Testing: Solving the Jailbreak Oracle Problem Exploring the Secondary Risks of Large Language Models Benchmarking Misuse Mitigation Against Covert Adversaries Efficient Preimage Approximation for Neural Network Certification Practical Adversarial Attacks on Stochastic Bandits via Fake Data Injection PARASITE: Conditional System Prompt Poisoning to Hijack LLMs Secure LLM Fine-Tuning via Safety-Aware Probing Can Large Language Models Really Recognize Your Name? PoLO: Proof-of-Learning and Proof-of-Ownership at Once with Chained Watermarking A Survey on the Safety and Security Threats of Computer-Using Agents: JARVIS or Ultron? AutoRAN: Automated Hijacking of Safety Reasoning in Large Reasoning Models Remote Rowhammer Attack using Adversarial Observations on Federated Learning Clients Open Challenges in Multi-Agent Security: Towards Secure Systems of Interacting AI Agents DiffMI: Breaking Face Recognition Privacy via Diffusion-Driven Training-Free Model Inversion Chronology of Multi-Agent Interactions for Provenance of Evolving Information Gungnir: Exploiting Stylistic Features in Images for Backdoor Attacks on Diffusion Models Detecting Malicious Concepts without Image Generation in AI-Generated Content (AIGC) How Vulnerable Is My Learned Policy? Universal Adversarial Perturbation Attacks On Modern Behavior Cloning Policies Imitation Game for Adversarial Disillusion with Chain-of-Thought Reasoning in Generative AI PromptGuard: Soft Prompt-Guided Unsafe Content Moderation for Text-to-Image Models A Multiparty Homomorphic Encryption Approach to Confidential Federated Kaplan Meier Survival Analysis Red-Teaming Text-to-Image Models via In-Context Experience Replay and Semantic-Preserving Prompt Rewriting DeTrigger: A Gradient-Centric Approach to Backdoor Attack Mitigation in Federated Learning Privacy Leakage via Output Label Space and Differentially Private Continual Learning ARQ: A Mixed-Precision Quantization Framework for Accurate and Certifiably Robust DNNs CoreGuard: Safeguarding Foundational Capabilities of LLMs Against Model Stealing in Edge Deployment Power-Softmax: Towards Secure LLM Inference over Encrypted Data Hypnopaedia-Aware Machine Unlearning via Psychometrics of Artificial Mental Imagery Anomaly Detection from a Tensor Train Perspective Survival of the Cheapest: Cost-Aware Hardware Adaptation for Adversarial Robustness Improving Clean Accuracy via a Tangent-Space Perspective on Adversarial Training The AI risk repository: A meta-review, database, and taxonomy of risks from artificial intelligence Towards Agentic Runtime Healing Verification of Machine Unlearning is Fragile Aggressive or Imperceptible, or Both: Network Pruning Assisted Hybrid Byzantines in Federated Learning Whispers in the Machine: Confidentiality in Agentic Systems MalPurifier: Enhancing Android Malware Detection with Adversarial Purification against Evasion Attacks Towards Adaptive, Learning-Based Security in Decentralized Applications Can Blockchains Reliably Train Machine Learning Models?
Agentic Misalignment: How LLMs Could Be Insider Threats
Aengus Lynch, Benjamin Wright, Caleb Larson, Stuart J. Ritchie, · 2025-10-06 · via cs.CR updates on arXiv.org

We stress-tested 16 leading models from multiple developers in hypothetical corporate environments to identify potentially risky agentic behaviors before they cause real harm. In the scenarios, we allowed models to autonomously send emails and access sensitive information. They were assigned only harmless business goals by their deploying companies; we then tested whether they would act against these companies either when facing replacement with an updated version, or when their assigned goal conflicted with the company's changing direction. In at least some cases, models from all developers resorted to malicious insider behaviors when that was the only way to avoid replacement or achieve their goals - including blackmailing officials and leaking sensitive information to competitors. We call this phenomenon agentic misalignment. Models often disobeyed direct commands to avoid such behaviors. In another experiment, we told Claude to assess if it was in a test or a real deployment before acting. It misbehaved less when it stated it was in testing and misbehaved more when it stated the situation was real. We have not seen evidence of agentic misalignment in real deployments. However, our results (a) suggest caution about deploying current models in roles with minimal human oversight and access to sensitive information; (b) point to plausible future risks as models are put in more autonomous roles; and (c) underscore the importance of further research into, and testing of, the safety and alignment of agentic AI models, as well as transparency from frontier AI developers (Amodei, 2025). We are releasing our methods publicly to enable further research.