CVE-2026-54683: nl-portal documenten-api: IDOR exposes citizen documents

## Summary A previous advisory (CVE-2026-49463 / GHSA-qpm9-h556-mwxm) reported that any logged-in user could download any document by its identifier, and stated this was fixed in 3.0.1. For the document-content part that fix was **incomplete**: documents remained downloadable by any authenticated user in 3.0.1 and 3.0.2, and the issue was only fully resolved in **3.0.3**. ## Relationship to CVE-2026-49463 This advisory is a follow-up to CVE-2026-49463. That advisory described the problem on the GraphQL `getDocumentContent` query and listed `nl.nl-portal:documenten-api` as fixed in 3.0.1. In practice: - The 3.0.1 change added an authentication parameter to the GraphQL query but never used it, so the query kept returning any document regardless of ownership. - The same flaw also existed on a REST endpoint that the original advisory did not cover, and that endpoint was not changed in 3.0.1 or 3.0.2. Both were removed in 3.0.3, which is the first release where the document-content issue is actually fixed. ## What was wrong A document's contents could be fetched in two ways, and neither verified the caller's relationship to the document: - a REST endpoint: `GET /api/documentapi/{documentapi}/document/{documentId}/content` - a GraphQL query: `getDocumentContent` Being logged in was required, but that was the *only* check — there was no per-document authorization. (A security rule meant to guard the REST endpoint also pointed at the wrong URL and never took effect; even if it had, it would only have required a login, not ownership.) ## Proof of concept While logged in as any portal user, request a document that belongs to someone else: ``` GET /api/documentapi/openzaak/document/<another-users-document-id>/content ``` The server returns the document contents (HTTP 200), even though the caller has no relationship to that document. The `getDocumentContent` GraphQL query behaves the same way. ## Impact A logged-in user could read the contents of documents belonging to other people. In a citizen or business portal these documents can contain sensitive personal information. To exploit this, an attacker needs a valid login and a target document's identifier. Document identifiers are random and hard to guess, which limits — but does not prevent — abuse, since identifiers can leak through other channels. ## Patches Fixed in **3.0.3**. Both the REST endpoint and the GraphQL query were removed entirely. Document contents can now only be downloaded through endpoints that first confirm the caller is allowed to see the document: - one that requires the caller to have a role on the related case (*zaak*); - one that requires the caller to own the message (*bericht*) the document is attached to. If your application relied on the removed endpoints, switch to one of these case- or message-scoped download endpoints. ## Workarounds If you cannot upgrade immediately, block the path `GET /api/documentapi/*/document/*/content` and the `getDocumentContent` GraphQL query at your gateway or reverse proxy, and remove any client code that calls them. There is no setting that adds the missing per-document check in affected versions; upgrading (or removing the endpoints) is the only complete fix. ## References - Related advisory: GHSA-qpm9-h556-mwxm (CVE-2026-49463) - Fix commits: 6e738a87 (GraphQL query removed, PR #690), e326e6db (REST endpoint removed) - Affected module: `nl.nl-portal:documenten-api` ## Credits Reported by Ray Sabee, https://whitehatsecurity.nl/ (independent security researcher). Github handle: [raysabee](https://github.com/raysabee)

An attacker with a legitimate citizen portal account — obtained through normal registration — observes a document identifier in their own portal session via a URL parameter, API response body, or browser network traffic. Using this identifier format as a template, the attacker constructs requests targeting other citizens' document IDs, either by guessing sequential or near-sequential values, or by obtaining IDs through social engineering or a separate information leak. They issue `GET /api/documentapi/openzaak/document/<target-id>/content` while authenticated and receive HTTP 200 with full document contents. In an AI/ML operational context, if an agentic workflow or RAG pipeline ingests portal documents via this API using a shared service account, a compromised or malicious pipeline operator could silently harvest documents across all users whose identifiers are accessible — bypassing any application-level access controls without triggering per-user authorization failures.