What we are learning about robot safety, instinct and building physical AI. We credit the work we build on.
NVIDIA is putting safety at the centre of physical AI. Here is how we see our work fitting beside theirs, with credit where it is due.
Hardware roots of trust get broken by glitching power. What a Tesla jailbreak teaches us about the limits of any safety chip.
The robot-learning world already talks about System 1 and System 2. We think a fixed layer deserves its own word.
Jailbreaks work on robots too. An old idea from control engineering suggests where the safety layer belongs.