Imagen de Portada

Evaluating Control Protocols for Untrusted AI Agents

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Kutasov, Jon, Loughridge, Chloe, Sun, Yuqi, Sleight, Henry, Shlegeris, Buck, Tracy, Tyler, Benton, Joe
Formato:	Preprint
Publicado:	2025
Materias:	Artificial Intelligence
Acceso en línea:	Acceder al recurso
Etiquetas:	Agregar Etiqueta Sin Etiquetas, Sea el primero en etiquetar este registro!

Ejemplares similares

Optimizing AI Agent Attacks With Synthetic Data
por: Loughridge, Chloe, et al.
Publicado: (2025)

SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agents
por: Kutasov, Jonathan, et al.
Publicado: (2025)

Games for AI Control: Models of Safety Evaluations of AI Deployment Protocols
por: Griffin, Charlie, et al.
Publicado: (2024)

Polysemanticity and Capacity in Neural Networks
por: Scherlis, Adam, et al.
Publicado: (2022)

SLEIGHT-Bench: A Benchmark of Evasion Attacks Against Agent Monitors
por: Najt, Elle, et al.
Publicado: (2026)

dafny-annotator: AI-Assisted Verification of Dafny Programs
por: Poesia, Gabriel, et al.
Publicado: (2024)

Model Spec Midtraining: Improving How Alignment Training Generalizes
por: Li, Chloe, et al.
Publicado: (2026)

A sketch of an AI control safety case
por: Korbak, Tomek, et al.
Publicado: (2025)

How to evaluate control measures for LLM agents? A trajectory from today to superintelligence
por: Korbak, Tomek, et al.
Publicado: (2025)

Auditing Sabotage Bench: A Benchmark for Detecting and Fixing Research Sabotage in ML Codebases
por: Gan, Eric, et al.
Publicado: (2026)

Factor(U,T): Controlling Untrusted AI by Monitoring their Plans
por: Lip, Edward Lue Chee, et al.
Publicado: (2025)

Safe, Untrusted, "Proof-Carrying" AI Agents: toward the agentic lakehouse
por: Tagliabue, Jacopo, et al.
Publicado: (2025)

BashArena: A Control Setting for Highly Privileged AI Agents
por: Kaufman, Adam, et al.
Publicado: (2025)

Subversion Strategy Eval: Can language models statelessly strategize to subvert control protocols?
por: Mallen, Alex, et al.
Publicado: (2024)

Language models are better than humans at next-token prediction
por: Shlegeris, Buck, et al.
Publicado: (2022)

Adaptive Deployment of Untrusted LLMs Reduces Distributed Threats
por: Wen, Jiaxin, et al.
Publicado: (2024)

Efficiently Aligning Language Models with Online Natural Language Feedback
por: Ye, Christine, et al.
Publicado: (2026)

Sabotage Evaluations for Frontier Models
por: Benton, Joe, et al.
Publicado: (2024)

Ctrl-Z: Controlling AI Agents via Resampling
por: Bhatt, Aryan, et al.
Publicado: (2025)

AI Organizations are More Effective but Less Aligned than Individual Agents
por: Shen, Judy Hanwen, et al.
Publicado: (2026)

The Hot Mess of AI: How Does Misalignment Scale With Model Intelligence and Task Complexity?
por: Hägele, Alexander, et al.
Publicado: (2026)

Singular Vectors of Attention Heads Align with Features
por: Franco, Gabriel, et al.
Publicado: (2026)

Structured Security Auditing and Robustness Enhancement for Untrusted Agent Skills
por: Lv, Lijia, et al.
Publicado: (2026)

All Code, No Thought: Current Language Models Struggle to Reason in Ciphered Language
por: Guo, Shiyuan, et al.
Publicado: (2025)

The LLM Has Left The Chat: Evidence of Bail Preferences in Large Language Models
por: Ensign, Danielle, et al.
Publicado: (2025)

Attack Selection Reduces Safety in Concentrated AI Control Settings against Trusted Monitoring
por: Schaeffer, Joachim, et al.
Publicado: (2026)

A Survey of AI Agent Protocols
por: Yang, Yingxuan, et al.
Publicado: (2025)

LinuxArena: A Control Setting for AI Agents in Live Production Software Environments
por: Tracy, Tyler, et al.
Publicado: (2026)

Bypassing AI Control Protocols via Agent-as-a-Proxy Attacks
por: Isbarov, Jafar, et al.
Publicado: (2026)

Agent Control Protocol: Admission Control for Agent Actions
por: Fernandez, Marcelo
Publicado: (2026)

MonitoringBench: Semi-Automated Red-Teaming for Agent Monitoring
por: Jotautaitė, Monika, et al.
Publicado: (2026)

PragLocker: Protecting Agent Intellectual Property in Untrusted Deployments via Non-Portable Prompts
por: Li, Qinfeng, et al.
Publicado: (2026)

Stress-Testing Model Specs Reveals Character Differences among Language Models
por: Zhang, Jifan, et al.
Publicado: (2025)

Data-Centric Interpretability for LLM-based Multi-Agent Reinforcement Learning
por: Yan, John, et al.
Publicado: (2026)

Deep Research Bench: Evaluating AI Web Research Agents
por: FutureSearch, et al.
Publicado: (2025)

From Prompts to Protocols: An AI Agent for Laboratory Automation
por: Angelopoulos, Angelos, et al.
Publicado: (2026)

Inverse Scaling in Test-Time Compute
por: Gema, Aryo Pradipta, et al.
Publicado: (2025)

Generative AI in clinical practice: novel qualitative evidence of risk and responsible use of Google's NotebookLM
por: Reuter, Max, et al.
Publicado: (2025)

KGCE: Knowledge-Augmented Dual-Graph Evaluator for Cross-Platform Educational Agent Benchmarking with Multimodal Language Models
por: Liu, Zixian, et al.
Publicado: (2026)

A survey of agent interoperability protocols: Model Context Protocol (MCP), Agent Communication Protocol (ACP), Agent-to-Agent Protocol (A2A), and Agent Network Protocol (ANP)
por: Ehtesham, Abul, et al.
Publicado: (2025)