If you switch AI providers tomorrow, what happens to your data, your context and the work your agents have already done?
That question stood out at the Sovereign AI sessions during SF Tech Week. Molly Mackinlay of FilOz gave a great presentation on Sovereign AI Needs a Sovereign Cloud.
The event also included separate talks and demonstrations on agent assets, infrastructure verification and Private by Design. These were different contributions to the broader conversation.
Ask what you can take with you
Ownership of AI memory brought a practical set of questions into focus: where does the data live, who can access it, what is retained and can we take our context with us? For me, the important point was to ask about the usable work left behind when a team changes tools.
Distinguish files from full agent memory
A demonstration showed one AI creating a file, storing it through Filecoin and IPFS and another retrieving it to continue the work. Moving files is a useful step. It does not, by itself, transfer an agent’s full memory, identity or permissions.
Ask what infrastructure evidence establishes
A presentation explored signed evidence that an AI system meets an approved security policy as new servers join. The questions I took away were what the evidence covers, who checks it and what happens when a server fails the check or access must be revoked.
Look at protection during processing
The Private by Design discussion explored confidential computing and encrypted AI. I want to understand who can access sensitive information during processing and how the claimed protections can be verified. The name of a technology alone is not a complete answer to those questions.
My takeaway
Evaluate AI tools beyond the quality of their answers. Ask whether you can export data, context and work in a usable format, what remains behind when you change providers, and how access and infrastructure protections can be verified.
Thank you to Molly Mackinlay for a great presentation and to the speakers and organizers for the talks and demonstrations.
Questions from this recap
Does portable file storage transfer an agent’s full memory?
No. The demonstration showed a file being created, stored and retrieved. That is useful portability, but full memory, identity and permissions are separate questions.
What should founders ask about sovereign AI?
Ask where data lives, who can access it, what is retained, whether context can be exported and what evidence supports the infrastructure’s protections.
Explore further
If you changed providers tomorrow, which parts of your team’s AI work could you actually carry forward?