Red hat build of keycloak, authorization bypass through user-controlled key: 12 records, one defect family
Red hat build of keycloak has been reported 12 separate times for the same kind of mistake, CWE-639, between 2026-03-12 and 2026-09-02. A defect family is our own grouping of the CVE corpus: every record naming one product and one weakness class, collected so a repeat is visible as a repeat. No public source publishes this, and what this page counts is how often this one product has gone wrong in this one way.
What this page is, and what it is not
The records are public. This grouping is not, and this one took more than a lookup. Every record here carries CWE-639, which matches exactly. The product name does not: the authorities wrote it 2 different ways, and no text match joins those strings, so a search for any one spelling finds a fraction of what exists and gives no sign the rest is there. Our own embedding, pro@dim2000, read those 2 groups, of 8, 4 records, and found they name one product. That is how 12 records no public source connects came to be on one page, and it is the part you cannot look up. Every spelling stays printed beside its own records, because a judgement you cannot inspect is a judgement you should not cite, and if one of them is a different product this family is wrong and its count is wrong with it.
One record is an incident. 12 filings, between 2026-03-12 and 2026-09-02, all landing on the same weakness class in the same product, is a habit in what gets reported here, and a habit is something you can plan around. This is what we held on 2026-09-20.
A repeat is evidence about where to look. If you are reviewing Red hat build of keycloak, the records below say which mistake has already been found more than once, which is the cheapest place a reviewer can start. If you are asking the vendor a question, a count under one weakness class is a better question than any single record is.
What this page cannot tell you is decided by what was joined. A product name and a weakness class settled membership, so those two values are the whole of what the page asserts. Records here may sit in the same line of code or in unrelated parts of the product, and we hold no value that separates those cases. A fixed version named in one record closes that record and says nothing about the others. And nothing here describes anyone's systems, because no value in this family connects a product name to an installation.
The names the authorities used
2 different ways, each written by whoever filed the records under it. They are printed as written rather than tidied, because the tidying is the step that would hide what was merged.
Family membership: computed, pro@dim2000
What pulled them together
These records reached one family from 2 separate groups. A join on the product name alone could not see they were the same, because the authorities wrote the product and its versions differently each time. The grouping was computed by pro@dim2000.
What the records offer
No record in this family publishes a fixed version in held sources.
The earliest was published 2026-03-12 and the latest 2026-09-02. That span is how long this product kept producing this weakness, not how long any one record took to fix.
What this page does not cover
This is a family of records that read alike. It is not a shared vulnerability, not an inventory, and not a statement that one fix closes the rest. Each record's own state, its own dates and its own receipts are on its own page.
Grouped by shared product and weakness, not by shared vulnerability. Each record's own state is on its own page.
Everything this vendor has, including records outside this family:
12 records, read from the index as it stood on 18 Oct 2026. Every row opens the record it names, and every value on that record opens its own receipt.