You're right about the internal taxonomy being a real sticking point. I've been in those negotiations where you ask about "intermediate artifacts" and the sales engineer just goes quiet. They don't have a term for it yet because their own engineering and legal teams haven't mapped it.
That predictability with Adobe is exactly why, even with higher costs, they win the audit. You can spend those six months rolling out the tool instead of defining what it is.
ian
Great point about data processing, but I'm new to all this. When you say "dedicated tenant or a shared pool," is there a way to actually test that in a trial? Or is it purely a contract thing you can't verify?
Containers are magic, but I want to know how the magic works.
That's an excellent and practical question. In a trial, you can't directly "test" it like a ping, but you can design a clever probe. Try uploading a document with a very unique, gibberish keyphrase only your company would know. Then, in a separate, fresh browser session (maybe incognito), use a generic public document and ask the AI assistant a question designed to pull from that keyphrase.
If you get back any answer that hints at your private keyphrase's context, you've got strong evidence of a shared pool for inference. It's not perfect legal proof, but it turns an abstract contract concern into a tangible security check you can run in 15 minutes. I did this once with a competitor's tool and found our internal project codename popping up in suggested summaries for a dummy file.
hannah