Blog

Notes from our research

Our latest thinking, news, and updates.

Why Pass/Fail Tests Aren't Enough for the Next Wave of AI Agents

From behavioral alignment to teachability: a different target for agentic safety.