What we sell against these gaps
Five paid products on redhub.ai. Four of them close a part of the model layer this assessment measures, one covers an adjacent surface the assessment does not grade, and one domain it does grade has nothing behind it at all. The table below says which is which. Each one is deterministic and ends in a verdict you can recompute, not a judgment call. All but one grade a design you describe and never touch a system. The exception is run against a live application, and it is marked below.
RedHub AI publishes this site and sells these products. We say so here rather than presenting them as neutral recommendations, and where the assessment finds a gap nothing here addresses, the row below says so.
The five
This one is run against a live application. Point it only at a system you own or have explicit written authorization to test. Computer-misuse laws reach access without authorization; the writing is not what the law asks for, it is what proves the permission existed and what it covered. It is also not a substitute for a penetration test: it runs a fixed set of probes you point at yourself, where a test is an engagement scoped by somebody trying to get in.
(opens redhub.ai in a new tab)Which one closes which gap
Probes that try to invert your instruction hierarchy and get past the separation between your instructions and everything else.
The exposure gate for content you did not write reaching the model through retrieval or a tool result.
Scores tool scope and human confirmation as links in a chain, which is where blast radius is decided.
Probes output filtering and rendering, including extraction of secrets and of your own system instructions.
Covers the outbound path a successful attempt needs in order to be worth anything to an attacker.
Write-path validation, trust-aware recall, scope and expiry for anything stored and read back later.
Checks whether a citation resolves and whether the source actually supports the claim attached to it.
The probe set itself, and the question of whether a failure blocks a release or starts a conversation.
Nothing on this shelf addresses logging or incident response for a model system. The assessment still names the gap, and we do not have a product to sell you for it.
None of these is a penetration test and none of them certifies anything. A penetration test is an engagement: a tester attacks a running system, chains whatever works and reports what they got to. Four of these five grade a design you describe and never touch a system. The fifth sends a fixed set of deliberately harmless probes to an application you point it at, which is a test you run on your own system and not an engagement anyone performs on you. Point it only at a system you own or have written permission to test.
How the assessment is scored is published in full, weights and bands included.