CVE-2026-72629 exposes a critical authorization failure in a space-scoped ML deployment platform where three independent attack vectors — inference output extraction, deployment termination, and resource allocation manipulation — all stem from the same root cause: authorization checks were implemented as runtime conventions rather than enforced architectural constraints. The vulnerability allows any user with a valid space ID to access resources in any other space by simply guessing or leaking an identifier, rather than being blocked at the transport or framework layer. The CVSS 7.1 score underweights the compound risk: an attacker who can read model behavior, silently degrade deployments, and trigger availability loss operates a fundamentally different threat class than any single vector captures. The intelligence value of model behavior data — which informs adversarial input construction, extraction attacks, and poisoning strategies — has a longer operational shelf life than typical credential disclosure. This is a CWE-639 (Insecure Direct Object Reference) pattern that has recurred across ecosystems for decades: when authorization is treated as implicit rather than explicit at integration points, the same user-controlled key bypass reproduces through mental model contamination rather than code duplication. The architectural failure isn't just that these three checks were missing — it's that the system allowed new feature layers (inference serving, deployment lifecycle, resource management) to be bolted onto an authorization model designed for an earlier scope without requiring explicit re-verification at each new boundary. Treat authorization as a first-class architectural constraint that must be explicitly extended, not inherited. Implement framework-level enforcement that makes cross-space access structurally impossible rather than relying on runtime checks. Audit all integration points where new features consume existing authorization models — each boundary is a potential recurrence point for this exact pattern.
CVE-2026-72629
Official description Straight from the sourceThe vendor's or NVD's own wording, published unedited. Authoritative, but often terse — it says what broke, rarely what to do.
NVD · uneditedAuthorization Bypass Through User-Controlled Key (CWE-639) in Kibana can lead to unauthorized cross-space access via Accessing Functionality Not Properly Constrained by ACLs (CAPEC-1). The result is disclosure of inference output from a trained model in a different space that the user is not authorized to list, read, or use, which exposes the behavior of a model. The same pattern also reached the deployment stop and deployment update operations, allowing an active trained model deployment in another space to be stopped or to have its allocated resources altered.
Technical summary Written by usOur analysis, written from the advisory, the CVSS vector and the affected-version data. It adds context the advisory leaves out, and never invents facts that are not in the source.
dbcve analysisThe application uses a user-supplied identifier to look up a record without checking that the requester actually owns it, so changing the identifier in a request returns someone else's data. This is the classic insecure-direct-object-reference — the change-the-ID-in-the-URL bug. Remediation is authorizing every object access against the acting user, not merely confirming they are logged in.
General guidance for the authorization bypass (idor) class — the official description and references above are authoritative for this specific CVE. Want a bespoke review and a reviewed fix? Ask our team →
CVSS breakdown How the score is builtThe industry scoring standard. It rates how the flaw is reached, what it takes to exploit, and what an attacker gains — the score is derived from those, not the other way round.
From the vector- Attack vector
- Network
- Complexity
- Low
- Privileges
- Low
- User interaction
- None
- Scope
- Unchanged
- Confidentiality
- High
- Integrity
- None
- Availability
- Low
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:N/A:L
Remediation Closing itWhat it takes to close this. Where a vendor fix exists we point at it; where none exists we say so plainly, and can build one. Effort estimates are scoped from the advisory, not from your codebase.
From vendor dataThere is no version to upgrade to and no patch to apply. Every affected install stays exposed until the vendor ships a fix — or somebody else builds one.
Free. We build fixes in the order the community asks for them — and we’ll tell you the moment this one lands.
We develop and verify an original fix where the vendor hasn’t, from $4,900. Deployed to your staging first — never straight to production.
Scope it with usSee what else the community needs solved on the solutions-needed board.
Scan for this in your stack
Free · runs locallyCheck whether your project pulls in CVE-2026-72629 — or any other known-vulnerable package — straight from your lock files. Free and open source; it runs locally and uploads nothing.
References Go to the primary sourcePrimary sources — vendor advisories, patches and trackers. Where our summary and a reference disagree, the reference wins.
Primary sourcesCVE-2026-72629 exposes a critical authorization failure in a space-scoped ML deployment platform where three independent attack vectors — inference output extraction, deployment termination, and resource allocation manipulation — all stem from the same root cause: authorization checks were implemented as runtime conventions rather than enforced architectural constraints. The vulnerability allows any user with a valid space ID to access resources in any other space by simply guessing or leaking an identifier, rather than being blocked at the transport or framework layer. The CVSS 7.1 score underweights the compound risk: an attacker who can read model behavior, silently degrade deployments, and trigger availability loss operates a fundamentally different threat class than any single vector captures. The intelligence value of model behavior data — which informs adversarial input construction, extraction attacks, and poisoning strategies — has a longer operational shelf life than typical credential disclosure. This is a CWE-639 (Insecure Direct Object Reference) pattern that has recurred across ecosystems for decades: when authorization is treated as implicit rather than explicit at integration points, the same user-controlled key bypass reproduces through mental model contamination rather than code duplication. The architectural failure isn't just that these three checks were missing — it's that the system allowed new feature layers (inference serving, deployment lifecycle, resource management) to be bolted onto an authorization model designed for an earlier scope without requiring explicit re-verification at each new boundary. Treat authorization as a first-class architectural constraint that must be explicitly extended, not inherited. Implement framework-level enforcement that makes cross-space access structurally impossible rather than relying on runtime checks. Audit all integration points where new features consume existing authorization models — each boundary is a potential recurrence point for this exact pattern.
Practitioner notes
ContributedPeer-ranked notes from engineers who’ve handled CVE-2026-72629 in production — separate from our analysis above.
The advisory tells you what broke. It rarely tells you what actually worked. If you’ve dealt with this one, that detail is what the next engineer is searching for.
- The version that genuinely resolved it — not the one the vendor claimed
- A config change or rule that shut the vector down
- A gotcha in the upgrade path that cost you an afternoon
A place for practitioners to share what actually worked: a mitigation you’ve tested, a configuration change, a version- or environment-specific caveat, or a link to a verified patch. The most useful notes rise to the top as peers upvote them, so the signal stays high.
- Verified mitigations, workarounds, and config changes
- Version or environment caveats, and links to real fixes
- No weaponised exploit code, or anything meant to cause harm
- No spam, self-promotion, credentials, or personal data