What we sell against these gaps
6 paid products on redhub.ai against eight domains, which leaves 2 with nothing behind them. Those 2 are named below, not quietly omitted, and one of them has a product on the shelf that nearly fits and is not being offered, which is also named. The domains came out of what an organization is actually exposed to; what happens to be on the shelf had no vote in which of them exist.
None of them scans, connects to or tests a system of yours. All 6 are structured procedures over a description you write down, and each ends in a verdict you can recompute, not a judgment call. Three of them are where the fatal absences on your result come from, each gating on the same thing about the one product it grades.
RedHub AI publishes this site and sells these products. We say so here rather than presenting them as neutral recommendations.
None of these is a penetration test and none of them certifies anything. Buying one changes nothing about what your organization is exposed to until somebody does the work it describes.
The 6
Which one closes which gap
Scores an AI vendor on what its terms actually commit to rather than on what its security page says. Two of its controls are disqualifiers, not weighted inputs: a written commitment not to train on your data, and a working deletion-and-exit clause. Either one absent forces RED FLAG whatever the rest scores, which is where the first of the three fatal controls on your result comes from.
Turns an honest list of the AI tools in use into a triage, grading each on sanction, what data it touches, whether it runs on a personal account and whether it trains on your data, and ending in a verdict per tool and one for the estate. You supply the list, by amnesty: it scans nothing and finds nothing on its own, and its own page says so. That is the design, because the tools people would not volunteer are the ones a scan is least likely to see.
Grades the controls standing between a convincing request and money moving or access changing. Its gate is one of the three on your result, and it is a pair rather than a single control: if neither an out-of-band callback nor a second approver is fully in place, one person can still push an urgent request through, and that reads EVIDENCE GAP regardless of the score. Bringing either one to present releases it, and this assessment reads it the same way.
Runs practical exercises with a team and reads the weakest drill, not the average, returning DRILLED, UNEVEN or RAW. Those are the three words on your own result, and they are borrowed from here. It also treats one unsafe reflex on regulated data as the whole exposure, whatever else scores well.
Grades whether you would actually recover from an outage or a ransomware event rather than whether you believe you would. Its gate is another of the three on your result: either a tested restore or an AI fallback reading zero forces WOULD NOT RECOVER, because knowing a backup exists is not knowing it recovers in time.
The go or no-go before something built quickly goes out, on a weighted score with two fatal controls over it: secrets out of the client bundle and repository, and a server-side check of who is asking on every sensitive route. Either absent forces DO NOT SHIP at any score, because one is a published credential and the other is a door with a sign on it.
The 2 we sell nothing against. Both are still asked, and both still carry weight in the score, because an assessment that only asked about the things we sell would be a different kind of document.
What you granted it. Nothing here closes this one. The nearest product grades the connectors and grants an agent holds, and what this domain asks about is the grant a person made to a SaaS tool by clicking through a consent screen: a different object, with a different owner, usually in a different admin console. Pointing you at it would be selling the nearest thing. The step that costs nothing is opening the third-party access page of whatever identity provider you use and reading the list.
Being convincingly faked. Nothing here closes this one either, and nothing honest could. Whether somebody can produce a convincing voice or video of a named person here is a fact about what is public about them, not about anything you can configure, and we do not sell detection. What is buyable is the control on the other side of it, which is the verification domain above.