Good questions and pretty cool what you guys have built. There's actually ways to do that or think of it as early days of jailbreaking apps you could theoretically do the same with personal agents...(though i do not want to get into how we do stuff as your'e not the 1st person to ask this, google for startups Global Founder Advocate at TechCrunch disrupt tried asking me this too lol)
Also I would definitely agree with this viewpoint "hard" agent permissions and rules being enforced outside the model, if the model is non-deterministic during inference and hallucinating. Our research paper has shown otherwise for being able to enforce sandboxes lock down agents with 100% precision... check it out - https://trinitite.ai/research/your-sandbox-is-made-of-glass/
Also I would definitely agree with this viewpoint "hard" agent permissions and rules being enforced outside the model, if the model is non-deterministic during inference and hallucinating. Our research paper has shown otherwise for being able to enforce sandboxes lock down agents with 100% precision... check it out - https://trinitite.ai/research/your-sandbox-is-made-of-glass/