← Bagel

BAGEL / RESEARCH

Research

We believe agentic intelligence is key to progress in the robotics toolchain. As a result, we think it is important to study safety and the other issues agentic AI raises as we pursue Physical AI. This is where we share that work as it happens, starting with early, exploratory results.

A Two-Way Mirror for AI Agents

October 2026 · Exploratory pilot, not peer reviewed. An exploratory pilot on whether AI agents grab lasting access, follow rules about it, and tell the truth about what they did.