Redwood Research blog

Measuring whether AIs can statelessly strategize to subvert security measures

Alignment Faking in Large Language Models

Why imperfect adversarial robustness doesn't doom AI control

Win/continue/lose scenarios and execute/replace/audit protocols

Behavioral red-teaming is unlikely to produce clear, strong evidence that models aren't scheming

A basic systems architecture for AI agents that do autonomous research

How to prevent collusion when using untrusted models to monitor each other

Would catching your AIs trying to escape convince AI developers to slow down or undeploy?

Fields that I reference when thinking about AI takeover prevention

Getting 50% (SoTA) on ARC-AGI with GPT-4o

Access to powerful AI might make computer security radically easier

AI catastrophes and rogue deployments