How do you enforce AI policy and security across multiple teams, using multiple models, tools, clouds, operating systems, etc? This is a massive challenge, and it’s only getting harder. The foundation model providers have governance and policy, but it protects them not you. Open source and open weight models have supply chain risks, and who can prove lineage and provenance of where that model or its dependencies originated? Then there’s the SaaS tool’s embedded AI (that no one asks for) … how do you secure that!? This is where Jozu Agent Guard comes in. We solved for both use cases. Build your own agents on open weight models … we will help you secure the supply chain, with cryptographic signatures for everything in the project (data, model, weights, prompt, doc, code, etc) all versioned together for audit and roll back purposes. Deploy it into prod with Jozu Agent Guard contained in a microVM, with the policy inside … meaning there’s no SaaS dependency or fail open … lose connectivity or operate air-gapped, all is good. Then there is use case two, desktop agent apps like Claude and ChatGPT. Agent Guard works with those as well. It runs behind the scenes, configured by your MDM tool or per device … the Agent doesn’t know it’s there, and it lets you enforce granular policy, and see when a user tries to violate that policy. Agents are becoming more engrained in our workflows, securing them and tying them to policy is what allows us to use them at their full potential.