This research was completed for London AI Safety Research (LASR) Labs 2025 by Jennifer Za, Julija Bainiaskina, Nikita Ostrovsky and Tanush Chopra. The team was supervised by Victoria Krakovna (Google DeepMind). Find out more about the programme and express interest in upcoming iterations here. Introduction Many proposals for controlling misaligned...
This post is an organizational update from Georgia Tech’s AI Safety Initiative (AISI) and roughly represents our collective view. In this post, we share lessons & takes from the 2024-25 academic year, describe what we’ve done, and detail our plans for the next academic year. Introduction Hey, we’re organizers of...
This research was completed for London AI Safety Research (LASR) Labs 2025 by Jennifer Za, Julija Bainiaskina, Nikita Ostrovsky and Tanush Chopra. The team was supervised by Victoria Krakovna (Google DeepMind). Find out more about the programme and express interest in upcoming iterations here.
Introduction
Many proposals for controlling misaligned LLM agents rely on using a second LLM as a monitor [1]. A concerning vulnerability of this setup is that agents may persuade monitors to approve undesirable actions.
Our research investigates the capability of frontier models to persuade monitors to approve actions that are against the monitor’s policy, and how this changes when the monitor has access to the agent’s chain-of-thought (CoT) [2]. We open source... (read 969 more words →)
This post is an organizational update from Georgia Tech’s AI Safety Initiative (AISI) and roughly represents our collective view. In this post, we share lessons & takes from the 2024-25 academic year, describe what we’ve done, and detail our plans for the next academic year.
Hey, we’re organizers of Georgia Tech’s AI Safety Initiative (AISI), thanks for dropping by! The purpose of this post is to document and share our activities with fellow student AI safety organizations. Feel free to skip to sections as you need. We welcome all constructive discussion! Put any further questions, feedback, or disagreements in the comments and one of us will certainly respond. A brief outline: