A trust company has to go first

There's an asymmetry in how vendors talk to compliance officers, and once you see it you can't unsee it. Every vendor will tell you, at length, what their product finds, catches, covers, protects. Almost none will hand you the other list — the one that says what their product misses, where it's blind, what it cannot see even in principle. That second list exists for every product ever built. The question is only whether the vendor gives it to you, or lets you discover it at the worst possible moment: in front of an examiner, when the gap you didn't know about becomes the gap you now have to explain.

We sell AI vendor-risk assessments to people whose entire job is being accountable for what they can prove. For a buyer like that, the missing list isn't a footnote. It's the more important document. So this piece is about a decision we made early and want to be explicit about: we publish our coverage boundary. Not buried in a terms page — in the report itself, on the website, in the sales conversation. A trust company doesn't get to ask for candor from vendors while being cagey about its own edges. It has to go first.

Read-only means read-only

Start with the phrase we use most, because it's the clearest place our discipline shows.

Our scan is read-only. When we say that, we mean it as a hard boundary in two directions, and both directions cost us something to hold.

It means we don't change anything. We don't revoke a grant, quarantine a tool, or flip a setting. We read your OAuth grants and your vendor surface, and we hand you back what's there. The revoking is yours — because the authority to touch your production identity fabric should be yours, and a vendor that quietly holds write access to your Google Workspace or Microsoft Entra is exactly the kind of standing-access risk we're supposed to be helping you find. We won't become the finding.

And it means we're honest that read-only sees less than the always-on, deeply-integrated posture a bigger platform might sell you. We see access — the standing grants, the scopes, who authorized what. We don't see every payload that access was used to move. Read-only is a narrower lens, and we'd rather hold the narrow lens honestly than claim a wide one we'd have to walk back. The discipline of the phrase is the discipline of the brand.

The boundary, published

Here is the kind of thing we mean by publishing the boundary — the actual edges, stated the way we'd want a vendor to state theirs to us:

• We show access, not payloads. A grant tells you a tool can read your Drive. It does not tell you what it read, when, or what it did with it. That's a harder question, and for some tools it's unanswerable no matter who's asking — audit-log export for the ChatGPT product line, for instance, is gated to the Enterprise and Edu tiers with 30-day retention, so a team on ChatGPT Business can't produce those records at all. We find the door. We don't pretend to hand you the camera footage behind it.
• We're point-in-time. A scan is true the day it runs and starts drifting the day after, because your people grant new access every week. The honest version is re-run on a cadence, not solved once — and we'll say "solved once" is a fiction every time we're asked.
• We have blind spots the old tools were good at. We don't catch member data pasted into a personal AI account on someone's phone — no corporate grant, no identity, nothing to inventory. We don't catch AI features baked inside a SaaS product that don't use a separate connection. We don't watch outbound content. For those, your network and DLP controls still earn their keep, and we'll tell you so rather than imply we replaced them.

None of those admissions helps a sales deck. All of them help a compliance officer sleep, because the one thing worse than a coverage gap is a coverage gap you were encouraged to believe wasn't there.

Why candor is the product, not a flaw in it

It would be easy to read the above as a company apologizing for its limits. It's the opposite. In a trust category, the coverage boundary is the product — here's the mechanism, not the sentiment.

A compliance officer's job is to make defensible statements about risk to people with authority over them. The most dangerous thing you can hand that person is false completeness — a green dashboard, a "you're covered" that papers over a gap. When the gap surfaces at an exam, it's now their credibility on the line, not the vendor's, because they're the one who repeated the claim. A vendor that overstates coverage isn't giving a compliance officer confidence. It's transferring risk to them and calling it reassurance.

So the honest boundary is the actual deliverable. A finding you can defend — "here's what we inventoried, here's the stated cadence, here's explicitly what it doesn't cover" — is worth more than a finding that claims more than it can hold. This is the same logic that runs through everything we publish: a documented refusal is stronger than a vague reassurance; a program honest about its edges is one an examiner believes about its center (see What an examiner-ready AI evidence binder contains and Shadow AI is an access problem not a traffic problem). Candor isn't the tax we pay to be a trust brand. It's the thing we're actually selling. The scan is how we deliver it.

Going first, on purpose

"A trust company has to go first" is a standard we're trying to hold before anyone makes us. It means publishing the boundary before a customer trips over it. It means read-only staying read-only even when write access would make a demo more impressive. It means, when a prospect asks "does this catch everything?", the answer is a plain "no, and here's exactly what it doesn't" — in the sales call, not the fine print.

We don't think this is charity, and we won't pretend it's easy — there will be quarters where a competitor's more confident claim wins a deal ours was too honest to make. We're betting that in a market built for people who get audited, the vendor who states its limits out loud becomes the one they trust with the things that matter. If we ever stop going first — if the boundary starts moving to the fine print, if "read-only" gets an asterisk — hold us to this page. Publishing it is the point. It's a promise you're allowed to check.

Every Discover scan we produce carries its own coverage-limits statement — what the read-only scan saw, the cadence it's true on, and in plain language what it doesn't cover. That page isn't a disclaimer we bury. It's the part we're proudest of, because it's the part that makes the rest defensible.

Request your Discover scan — we run it with you, and we'll tell you what it misses before you have to ask. Want the boundary before the report? It's on Shadow AI is an access problem not a traffic problem and in the honest close of every playbook we publish.