AI Transparency and the Approval Question

Transparency is not a values slide. It is whether you can always answer three questions: who approved this, what did the AI assume, and where is the record?

Trust is not a feeling. It is something you can check.

A lot of AI products talk about “trust” and “transparency” as a tone: reassuring language, a friendly explanation, a badge on the pricing page. We think that is the wrong test. The real question is much simpler. When this thing takes an action on your behalf, can you actually verify what happened?

That is the standard we hold Gravitre to. Not a promise. A mechanism you can go check yourself.

Does every AI action need approval?

You might ask Gravitre to do something directly in a chat message. You might be stepping through a guided task. Or you might have a full automated workflow running in the background. However the request comes in, it hits the exact same approval standard. There is no side door where a workflow gets to skip the rules a chat message would follow.

That is the bar we hold ourselves to. It should not matter which path a request comes through. If it needs approval, it waits for approval, every time.

Who approves AI actions, and what everyone sees

Approval is not a setting someone can quietly switch off for convenience. It is built into how every write action works:

• If you have approval rights and your company's policy allows it, you will see a confirmation step right in the moment, and can approve it yourself.
• If you do not have approval rights, your request does not just vanish or run anyway. You will see a clear message that it has been sent for approval, and it lands in a queue an approver can review at Approvals.
• Your company controls all of this from one place: who can approve what, and for which kinds of actions, in Settings → Approvals.

For multi-step automated runs, each step is labeled up front as something that just reads information, or something that changes it and needs approval. That way you know exactly what is coming before it happens, not after something has already changed.

Real proof it worked, not just a checkmark

A green checkmark that does not lead anywhere is not proof of anything. When a write action succeeds, Gravitre shows you one of a few honest outcomes:

• Verified: a direct link to the record that was created or updated, so you can go look at it yourself.
• Completed, with a summary: a direct link is not available, but you still get a clear description of what happened.
• Failed: neither a link nor a clear result is available. We would rather tell you it failed than show you a success we cannot back up.

When Gravitre guesses, it tells you

Sometimes completing a task means filling in a small detail you did not explicitly provide, like a default name for something you asked to create. When that happens, Gravitre shows you an Assumptions note, separate from the summary of what it did, so you always know what came directly from you and what the AI filled in on its own.

This is still expanding to cover more situations over time, and we would rather be upfront that it is not everywhere yet than pretend every judgment call the AI makes is spelled out today.

Suggestions you can trust, not a black box

Gravitre also surfaces proactive suggestions on Intelligence, like recommending you connect a tool you have not set up yet, or pointing out a useful next step. Every one of these comes with a plain-language reason and the evidence behind it, so you are never just told to trust it.

And importantly, these suggestions are just that: suggestions. There is no button that lets a recommendation execute itself. A human decides.

Real citations, or an honest “here is where to look”

When Gravitre answers a question using your connected knowledge sources, you will see real, numbered citations you can click through to the source. If a search does not turn up a good match, we do not invent a citation to look complete. You will see a clear link to check your sources directly at Sources.

Is there an audit trail for every AI action?

Every approval and every action is logged, and you can review, filter, and export that history any time from your audit page. If you are investigating something after the fact, whether that is a change you did not expect or confirming a policy was followed, the record is there.

We will be upfront that this is an area we are continuing to strengthen. Some advanced audit features are tied to specific plans today, and we are working toward an even more unified view. But the core principle holds: nothing important happens without a trace you can go find.

What is next

We are expanding assumption labeling to cover more situations, broadening citation coverage, and tightening the audit experience even further. As always, we will tell you plainly what is shipped versus what is still on the way. That is the same discipline that shapes every claim on this page.

If you are also curious about how we think about measuring the actual impact of automation, our post on measuring what AI actually changes is a natural next read.