Its clear that having broad permissions leaves room for Agents to go awry. This example of task bound permission's successful defence is very interesting! Curious to see this being leveraged in production.

I've implemented Tenuo with Temporal recently and have seen prompt-injection fail outright on planted tests. Honestly super useful considering the Agents i'm building require trust to operate on client data without the failure of injection.