We are creating designs for open, tamper-proof compute hardware (e.g. GPUs) that can locally enforce policies like export controls or safety evaluations. This hardware could encode laws restricting certain usages of AI or requiring use of safety best-practices or enable servers that can remain secure even if a hacker has physical access to them. These capabilities could be accomplished if the hardware itself could check usage against policies, and disable itself in the event of a violation. We have hosted workshops and demonstrated prototypes in AI Security Forum and other venues; and also raised philanthropic seed funding for the effort.