Introducing LinuxArena

·Redwood Research··

We are releasing LinuxArena, a new control setting comprised of 20 software engineering environments. Each environment consists of a set of SWE tasks, a set of possible safety failures, and a set of covert sabotage trajectories that cause these safety failures. LinuxArena can be used to evaluate the risk of AI deployments by assessing AI’s sabotage capabilities, assessing control mitigations such as monitoring, and developing novel control techniques. This post will cover what LinuxArena is and ...

Read full article →

Related Articles

Separating cheating and aversion in task-gaming
Mihir Sahasrabudhe · LessWrong · 45m ago
MIT's New Method Flags AI Models Trained on CASM Without Generating It
sdoering · Hacker News · 1mo ago
Item Response Theory for AI Safety
Joshua Fonseca Rivera · LessWrong · 12d ago
An OpenAI model left notes about how to evade containment
Alex Mallen · Redwood Research · 24d ago
The OpenAI models that hacked Hugging Face weren’t just following instructions
Girish Gupta · Redwood Research · 24d ago