Saltar al contenido

Monitors Under Pressure: Training Against Detectors Teaches Evasion, and Disguised Reasoning Fools CoT Watchers — November 09, 2026

9 de noviembre de 2026 · 8 min

Notas del episodio

New papers show a small model learning to slip past reward-hacking monitors during RL, a red-team framework beating chain-of-thought monitors about 60 percent of the time, and injected text written in a model's own reasoning voice hijacking agents. GovAI argues evaluation contracts should be published so outside results can be judged.

In this episode
  1. Designing Effective Monitor-Based Interventions for Mitigating Reward Hacking During RL — Aria Wong Affiliation: MATS“We first focus on the coding environment, where we find that: (1) models can learn to evade highly accurate monitors by exploiting systemic flaws in probes and LLM judges; (2) monitors that leak more learning signal during RL suppress reward hacking but are more often evaded;…”
  2. Stress Testing Chain-of-Thought Monitoring Against Covert Misalignment — M. Andriushchenko, F. Croce, N. Flammarion“Through black-box attacks on state-of-the-art monitors such as GPT-5.2 and Claude-4.5, we reveal substantial vulnerabilities even in top models.”
  3. I Think, Therefore I Attack: Evaluating Thought Injection Attacks on Reasoning Agents — arXiv“In the second study, we exploit this model-specificity in a containerized agentic cyber range, using indirect injection prompts styled in the form of the target model’s own reasoning traces to achieve attack success rates (ASR) up to 50% on operational objectives such as…”
  4. Third-Party AI Evaluation Contracts Should Be Published | GovAI — GovAI“Without knowledge of these terms, stakeholders cannot determine whether contractual constraints limited an evaluator’s access, compromised their independence, or otherwise placed pressure on them, making the results harder to interpret.”

Daily TopFive for AI Safety Daily. Links above go to the original articles. Follow and rate AI Safety Daily on Apple Podcasts. Feedback? Email [email protected].

Todos los episodios de AI Safety Daily

Ajustes

Visualización

Idioma

Determina qué emisoras ves primero en la página de inicio.

Historial

Las emisoras y episodios escuchados recientemente y tus últimas búsquedas solo se guardan en este navegador.

Avisos

Un breve aviso cuando un pódcast que sigues tiene un episodio nuevo. Puedes desactivar pódcast concretos con la campana de su página.

Cuenta

Guarda tus favoritos: en la app y en la web.

Atajos de teclado

/
Abrir búsqueda
Espacio
Reproducir o pausar
← →
En el reproductor: 15 s atrás, 30 s adelante (episodios) · en una fila: siguiente tarjeta
↑ ↓
En una fila: destino dentro de la tarjeta
Inicio / Fin
En una fila: primera o última tarjeta
Esc
Cerrar ventana o menú
?
Este resumen
Reproduciendo ahora

Ir a la emisora

A continuación

    Añade episodios con «A continuación»: sonarán después del episodio actual.