Imagen de Portada

Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation

Fuente: arXiv

Guardado en:

Detalles Bibliográficos
Autores principales:	Baker, Bowen, Huizinga, Joost, Gao, Leo, Dou, Zehao, Guan, Melody Y., Madry, Aleksander, Zaremba, Wojciech, Pachocki, Jakub, Farhi, David
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

Monitoring Monitorability
por: Guan, Melody Y., et al.
Publicado: (2025)

Hodoscope: Unsupervised Monitoring for AI Misbehaviors
por: Zhong, Ziqian, et al.
Publicado: (2026)

Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety
por: Korbak, Tomek, et al.
Publicado: (2025)

Small-to-Large Generalization: Data Influences Models Consistently Across Scale
por: Khaddaj, Alaa, et al.
Publicado: (2025)

DsDm: Model-Aware Dataset Selection with Datamodels
por: Engstrom, Logan, et al.
Publicado: (2024)

Decomposing and Editing Predictions by Modeling Model Computation
por: Shah, Harshay, et al.
Publicado: (2024)

AutoMonitor-Bench: Evaluating the Reliability of LLM-Based Misbehavior Monitor
por: Yang, Shu, et al.
Publicado: (2026)

Can Reasoning Models Obfuscate Reasoning? Stress-Testing Chain-of-Thought Monitorability
por: Zolkowski, Artur, et al.
Publicado: (2025)

Ask Your Distribution Shift if Pre-Training is Right for You
por: Cohen-Wang, Benjamin, et al.
Publicado: (2024)

User Strategization and Trustworthy Algorithms
por: Cen, Sarah H., et al.
Publicado: (2023)

Do Large Language Model Benchmarks Test Reliability?
por: Vendrow, Joshua, et al.
Publicado: (2025)

Learning to Attribute with Attention
por: Cohen-Wang, Benjamin, et al.
Publicado: (2025)

De hand van Huizinga
por: Huizinga, Johan
Publicado: (2010)

Privatization, public investment, and capital income taxation / Harry Huizinga, Søren Bo Nielsen
por: Huizinga, Harry
Publicado: (1997)

Homo ludens / Johan Huizinga ; traductor Eugenio Imaz
por: Huizinga, Johan

El otoño de la edad media : estudios sobre la forma de la vida y del espíritu durante los siglos XIV y XV en Francia en los Países Bajos / Johan Huizinga ; versión española de José Gaos
por: Huizinga, Johan
Publicado: (1981)

El elemento estético de las representaciones históricas
por: Johan Huizinga
Publicado: (2005)

Existe uma metamorfose da História? Resposta à pergunta: como o presente se torna passado? (Berliner Tageblatt, 31 de maio de 1936)
por: Johan Huizinga
Publicado: (2015)

ContextCite: Attributing Model Generation to Context
por: Cohen-Wang, Benjamin, et al.
Publicado: (2024)

The modal theory of linear orders
por: Wołoszyn, Wojciech Aleksander
Publicado: (2026)

Modal group theory
por: Wołoszyn, Wojciech Aleksander
Publicado: (2026)

The modal theory of the category of sets
por: Wołoszyn, Wojciech Aleksander
Publicado: (2026)

Grzegorczyk Logic Unlocked
por: Wołoszyn, Wojciech Aleksander
Publicado: (2025)

Modal group theory: homomorphisms
por: Wołoszyn, Wojciech Aleksander
Publicado: (2026)

Atoms and Molecules as Quantum Attosecond Processors
por: Farhi, Asaf
Publicado: (2025)

On refinements of two-term Machin-like formulas
por: Farhi, Bakir
Publicado: (2026)

Expansion of a bivariate symmetric mean in the neighborhood of the first bisector
por: Farhi, Bakir
Publicado: (2025)

R2L: Reliable Reinforcement Learning: Guaranteed Return & Reliable Policies in Reinforcement Learning
por: Farhi, Nadir
Publicado: (2025)

A measure of intelligence of an approximation of a real number in a given model
por: Farhi, Bakir
Publicado: (2017)

$q$-analogues of sums of consecutive powers of natural numbers and extended Carlitz $q$-Bernoulli numbers and polynomials
por: Farhi, Bakir
Publicado: (2025)

New formulas involving Bernoulli and Stirling numbers of both kinds
por: Farhi, Bakir
Publicado: (2024)

Pilgrim interaction with services provided by the general presidency of Alharamain affairs
por: F. Farhi
Publicado: (2020)

Training Agents to Self-Report Misbehavior
por: Lee, Bruce W., et al.
Publicado: (2026)

Wink: Recovering from Misbehaviors in Coding Agents
por: Nanda, Rahul, et al.
Publicado: (2026)

Misalignment Bounty: Crowdsourcing AI Agent Misbehavior
por: Turtayev, Rustem, et al.
Publicado: (2025)

LLMScan: Causal Scan for LLM Misbehavior Detection
por: Zhang, Mengdi, et al.
Publicado: (2024)

Measuring Strategization in Recommendation: Users Adapt Their Behavior to Shape Future Content
por: Cen, Sarah H., et al.
Publicado: (2024)

Training on Documents About Monitoring Leads to CoT Obfuscation
por: Haskins, Reilly, et al.
Publicado: (2026)

Misbehavior Forecasting for Focused Autonomous Driving Systems Testing
por: Naziri, M M Abid, et al.
Publicado: (2025)

Communism – Legitimacy – Nationalism
por: Zaremba, Marcin
Publicado: (2024)