GHSA-J9FC-W3MR-X6MV
Vulnerability from github – Published: 2026-07-24 21:12 – Updated: 2026-08-12 18:57Summary
Budibase 3.39.19 (commit 03fbabae4) is affected by a privilege-escalation / missing-authorization flaw in the public role-assignment API. An app-scoped builder (a user who builds only specific apps — user.builder.apps = [appA], not a global builder or admin) can grant themselves builder access to ANY other app in the tenant, or assign themselves/any user an arbitrary data-plane role (e.g. ADMIN) in any app, by calling POST /api/public/v1/roles/assign. The endpoint only authorizes the two global flags (admin, builder); the per-app appBuilder and role:{appId,roleId} grant vectors are passed to the backend without any authorization check that the caller controls the target app. Reproduced in a local authorized lab using the verbatim authorization-gate and SDK logic.
This is an incomplete fix of the role-assignment hardening in commit d63d1d9054 ("Inline public user global role validation"), which only ever validated the global flags.
Details
The issue is caused by authorization being implemented as a flag-level allowlist (admin/builder) instead of validating the scope the caller is granting.
Relevant code paths:
packages/server/src/api/controllers/public/globalRoleValidation.ts—validateGlobalRoleUpdate(ctx, roleUpdate)only checksroleUpdate.admin(requiresisAdmin) androleUpdate.builder(requiresisGlobalBuilder). TheGlobalRoleUpdateinterface declares only{ builder?, admin? };appBuilderandroleare not referenced.packages/server/src/api/controllers/public/roles.ts—assign()doesconst { userIds, ...assignmentProps } = ctx.request.body; validateGlobalRoleUpdate(ctx, assignmentProps); await sdk.publicApi.roles.assign(userIds, assignmentProps). TheappBuilder/roleprops pass through unvalidated.packages/pro/src/sdk/publicApi/roles.ts—assign(): foropts.appBuilderit setsuser.builder = { apps: existing.concat([getProdWorkspaceID(opts.appBuilder.appId)]) }; foropts.roleit setsuser.roles[getProdWorkspaceID(opts.role.appId)] = opts.role.roleId. No check that the caller builds the target app, anduserIdsis an arbitrary list (bulkGet). The only gate is theisExpandedPublicApiEnabled()license check.packages/server/src/api/routes/public/index.ts—applyAdminRoutes(roleEndpoints)attaches onlymiddleware.builderOrAdmin(nopublicApi, noauthorized(PermissionType.USER, …)).packages/backend-core/src/middleware/builderOrAdmin.ts— with aworkspaceIdpresent, it only requiresisBuilder(ctx.user, workspaceId). The attacker setsx-budibase-app-idto their own app (appA), so the gate passes.packages/backend-core/src/middleware/builderOnly.ts— on the worker,POST /api/global/self/api_keyonly requireshasBuilderPermissions(ctx.user), which is true for app-scoped builders, so the attacker can self-issue a public API key.packages/shared-core/src/sdk/documents/users.ts—isGlobalBuilderis false for app-scoped builders (so the globalbuilderflag is correctly blocked), whileisBuilder(user, appA)andhasBuilderPermissions(user)are true.
Attack flow:
- Attacker is an app-scoped builder of
appAonly (no global builder/admin). POST /api/global/self/api_key(worker) — passesbuilderOnlyviahasBuilderPermissions→ attacker obtains a public API key.POST /api/public/v1/roles/assignwith headerx-budibase-app-id: <appA prod id>and body{"userIds":["<self>"],"appBuilder":{"appId":"<appB>"}}—builderOrAdminpasses (isBuilder(user, appA)),validateGlobalRoleUpdateignoresappBuilder, the SDK pushesappBintouser.builder.apps.- Attacker is now a builder of
appB(and, via therolevector, can set any data-role such asADMINin any app).
Security boundary crossed:
- Before: builder of
appAonly. - After: builder of
appB(and any other app) → read/modify all rows, read datasource configs and exfiltrate stored datasource credentials, edit automations (including thebash/executeScript/executeQuerysteps); plus arbitrary data-role assignment in any app. - Why disallowed: the per-app builder model is meant to isolate builders to their assigned apps; a builder of one app must not gain authority over apps they were never granted.
PoC
Environment:
- Budibase version:
3.39.19, commit03fbabae4 - Deployment: Business/Enterprise license required (
isExpandedPublicApiEnabled) - Attacker role: app-scoped builder (
user.builder.apps=[appA]), not global builder/admin - Mock services: none needed for the unit-level proof; HTTP PoC script provided for a licensed lab
Steps (unit-level proof, no license needed — verbatim auth-gate + SDK logic):
- Run the verification harness:
cd D:/CVE-Hunting/budibase-audit-output/lab
node harness/verify-privesc-authz.cjs
- Observed result (evidence:
evidence/lab-privesc-authz.log):
[PASS-VALIDATION] appBuilder:{appId:APP_B} -> NOT rejected
[PASS-VALIDATION] role:{appId:APP_B, roleId:'ADMIN'}-> NOT rejected
[BLOCKED 403] builder:true (global) -> Only global builders or admins ...
[BLOCKED 403] admin:true (global) -> Only global admins ...
after : builder={"apps":["app_A...","app_B..."]} roles={"app_B...":"ADMIN"}
isBuilder(attacker, APP_B) now = true <-- escalated to builder of APP_B
Steps (HTTP PoC for a licensed lab — poc/privesc-roles-assign.sh):
# 1) self-issue API key
curl -X POST http://localhost:10000/api/global/self/api_key -H "Cookie: <attacker session>" -d '{}'
# 2) escalate: grant self builder of appB
curl -X POST http://localhost:10000/api/public/v1/roles/assign \
-H "x-budibase-api-key: <KEY>" -H "x-budibase-app-id: <appA prod id>" \
-H "content-type: application/json" \
-d '{"userIds":["<self global id>"],"appBuilder":{"appId":"<appB>"}}'
- Expected result:
The request should be rejected (403) — an app-scoped builder must not be able to grant
itself builder/role access to an app it does not control. Instead it returns 200 and the
grant is applied.
Evidence files:
evidence/lab-privesc-authz.logpoc/verify-privesc-authz.cjs,poc/privesc-roles-assign.sh
Runtime limitation: full HTTP end-to-end requires a Business/Enterprise license (isExpandedPublicApiEnabled); no license was cracked. The authorization defect is proven at unit level with the verbatim validation + SDK logic, and the route→validation→SDK call chain was independently verified by source review.
Source PoC
Impact
An attacker holding an app-scoped builder role on a single app in a licensed (Business/Enterprise) tenant can:
- escalate to builder of every other app in the tenant (cross-app/workspace isolation bypass);
- read and modify all rows in those apps;
- read those apps' datasource configurations and exfiltrate stored datasource credentials;
- edit automations in those apps, including server-side execution steps;
- assign arbitrary data-plane roles (e.g.
ADMIN) in any app to any user, including themselves.
It does not grant the global admin/builder flags (those are validated), so it is not a direct global-admin takeover; but cross-app builder is effectively a tenant-wide app/data-plane compromise. No real secrets were accessed during testing; the lab used canary-only data.
{
"affected": [
{
"package": {
"ecosystem": "npm",
"name": "@budibase/server"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"last_affected": "3.38.1"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-73305"
],
"database_specific": {
"cwe_ids": [
"CWE-269",
"CWE-862",
"CWE-863"
],
"github_reviewed": true,
"github_reviewed_at": "2026-07-24T21:12:40Z",
"nvd_published_at": null,
"severity": "HIGH"
},
"details": "### Summary\n\nBudibase `3.39.19` (commit `03fbabae4`) is affected by a privilege-escalation / missing-authorization flaw in the public role-assignment API. An **app-scoped builder** (a user who builds only specific apps \u2014 `user.builder.apps = [appA]`, not a global builder or admin) can grant **themselves builder access to ANY other app in the tenant**, or assign themselves/any user an arbitrary data-plane role (e.g. `ADMIN`) in any app, by calling `POST /api/public/v1/roles/assign`. The endpoint only authorizes the two *global* flags (`admin`, `builder`); the per-app `appBuilder` and `role:{appId,roleId}` grant vectors are passed to the backend **without any authorization check that the caller controls the target app**. Reproduced in a local authorized lab using the verbatim authorization-gate and SDK logic.\n\nThis is an incomplete fix of the role-assignment hardening in commit `d63d1d9054` (\"Inline public user global role validation\"), which only ever validated the global flags.\n\n### Details\n\nThe issue is caused by authorization being implemented as a **flag-level allowlist** (`admin`/`builder`) instead of validating the *scope* the caller is granting.\n\nRelevant code paths:\n\n- `packages/server/src/api/controllers/public/globalRoleValidation.ts` \u2014 `validateGlobalRoleUpdate(ctx, roleUpdate)` only checks `roleUpdate.admin` (requires `isAdmin`) and `roleUpdate.builder` (requires `isGlobalBuilder`). The `GlobalRoleUpdate` interface declares only `{ builder?, admin? }`; `appBuilder` and `role` are not referenced.\n- `packages/server/src/api/controllers/public/roles.ts` \u2014 `assign()` does `const { userIds, ...assignmentProps } = ctx.request.body; validateGlobalRoleUpdate(ctx, assignmentProps); await sdk.publicApi.roles.assign(userIds, assignmentProps)`. The `appBuilder`/`role` props pass through unvalidated.\n- `packages/pro/src/sdk/publicApi/roles.ts` \u2014 `assign()`: for `opts.appBuilder` it sets `user.builder = { apps: existing.concat([getProdWorkspaceID(opts.appBuilder.appId)]) }`; for `opts.role` it sets `user.roles[getProdWorkspaceID(opts.role.appId)] = opts.role.roleId`. No check that the caller builds the target app, and `userIds` is an arbitrary list (`bulkGet`). The only gate is the `isExpandedPublicApiEnabled()` license check.\n- `packages/server/src/api/routes/public/index.ts` \u2014 `applyAdminRoutes(roleEndpoints)` attaches **only** `middleware.builderOrAdmin` (no `publicApi`, no `authorized(PermissionType.USER, \u2026)`).\n- `packages/backend-core/src/middleware/builderOrAdmin.ts` \u2014 with a `workspaceId` present, it only requires `isBuilder(ctx.user, workspaceId)`. The attacker sets `x-budibase-app-id` to **their own** app (appA), so the gate passes.\n- `packages/backend-core/src/middleware/builderOnly.ts` \u2014 on the worker, `POST /api/global/self/api_key` only requires `hasBuilderPermissions(ctx.user)`, which is true for app-scoped builders, so the attacker can self-issue a public API key.\n- `packages/shared-core/src/sdk/documents/users.ts` \u2014 `isGlobalBuilder` is false for app-scoped builders (so the global `builder` flag is correctly blocked), while `isBuilder(user, appA)` and `hasBuilderPermissions(user)` are true.\n\nAttack flow:\n\n1. Attacker is an app-scoped builder of `appA` only (no global builder/admin).\n2. `POST /api/global/self/api_key` (worker) \u2014 passes `builderOnly` via `hasBuilderPermissions` \u2192 attacker obtains a public API key.\n3. `POST /api/public/v1/roles/assign` with header `x-budibase-app-id: \u003cappA prod id\u003e` and body `{\"userIds\":[\"\u003cself\u003e\"],\"appBuilder\":{\"appId\":\"\u003cappB\u003e\"}}` \u2014 `builderOrAdmin` passes (`isBuilder(user, appA)`), `validateGlobalRoleUpdate` ignores `appBuilder`, the SDK pushes `appB` into `user.builder.apps`.\n4. Attacker is now a builder of `appB` (and, via the `role` vector, can set any data-role such as `ADMIN` in any app).\n\nSecurity boundary crossed:\n\n- **Before:** builder of `appA` only.\n- **After:** builder of `appB` (and any other app) \u2192 read/modify all rows, read datasource configs and exfiltrate stored datasource credentials, edit automations (including the `bash`/`executeScript`/`executeQuery` steps); plus arbitrary data-role assignment in any app.\n- **Why disallowed:** the per-app builder model is meant to isolate builders to their assigned apps; a builder of one app must not gain authority over apps they were never granted.\n\n### PoC\n\nEnvironment:\n\n- Budibase version: `3.39.19`, commit `03fbabae4`\n- Deployment: Business/Enterprise license required (`isExpandedPublicApiEnabled`)\n- Attacker role: app-scoped builder (`user.builder.apps=[appA]`), not global builder/admin\n- Mock services: none needed for the unit-level proof; HTTP PoC script provided for a licensed lab\n\nSteps (unit-level proof, no license needed \u2014 verbatim auth-gate + SDK logic):\n\n1. Run the verification harness:\n\n```bash\ncd D:/CVE-Hunting/budibase-audit-output/lab\nnode harness/verify-privesc-authz.cjs\n```\n\n2. Observed result (evidence: `evidence/lab-privesc-authz.log`):\n\n```text\n[PASS-VALIDATION] appBuilder:{appId:APP_B} -\u003e NOT rejected\n[PASS-VALIDATION] role:{appId:APP_B, roleId:\u0027ADMIN\u0027}-\u003e NOT rejected\n[BLOCKED 403] builder:true (global) -\u003e Only global builders or admins ...\n[BLOCKED 403] admin:true (global) -\u003e Only global admins ...\nafter : builder={\"apps\":[\"app_A...\",\"app_B...\"]} roles={\"app_B...\":\"ADMIN\"}\nisBuilder(attacker, APP_B) now = true \u003c-- escalated to builder of APP_B\n```\n\nSteps (HTTP PoC for a licensed lab \u2014 `poc/privesc-roles-assign.sh`):\n\n```bash\n# 1) self-issue API key\ncurl -X POST http://localhost:10000/api/global/self/api_key -H \"Cookie: \u003cattacker session\u003e\" -d \u0027{}\u0027\n# 2) escalate: grant self builder of appB\ncurl -X POST http://localhost:10000/api/public/v1/roles/assign \\\n -H \"x-budibase-api-key: \u003cKEY\u003e\" -H \"x-budibase-app-id: \u003cappA prod id\u003e\" \\\n -H \"content-type: application/json\" \\\n -d \u0027{\"userIds\":[\"\u003cself global id\u003e\"],\"appBuilder\":{\"appId\":\"\u003cappB\u003e\"}}\u0027\n```\n\n3. Expected result:\n\n```text\nThe request should be rejected (403) \u2014 an app-scoped builder must not be able to grant\nitself builder/role access to an app it does not control. Instead it returns 200 and the\ngrant is applied.\n```\n\nEvidence files:\n\n- `evidence/lab-privesc-authz.log`\n- `poc/verify-privesc-authz.cjs`, `poc/privesc-roles-assign.sh`\n\n**Runtime limitation:** full HTTP end-to-end requires a Business/Enterprise license (`isExpandedPublicApiEnabled`); no license was cracked. The authorization defect is proven at unit level with the verbatim validation + SDK logic, and the route\u2192validation\u2192SDK call chain was independently verified by source review.\n\n### Source PoC\n\n[Budibase-GHSA-Reports.zip](https://github.com/user-attachments/files/29161435/Budibase-GHSA-Reports.zip)\n\n### Impact\n\nAn attacker holding an **app-scoped builder** role on a single app in a licensed (Business/Enterprise) tenant can:\n\n* escalate to builder of **every other app** in the tenant (cross-app/workspace isolation bypass);\n* read and modify all rows in those apps;\n* read those apps\u0027 datasource configurations and **exfiltrate stored datasource credentials**;\n* edit automations in those apps, including server-side execution steps;\n* assign arbitrary data-plane roles (e.g. `ADMIN`) in any app to any user, including themselves.\n\nIt does **not** grant the global `admin`/`builder` flags (those are validated), so it is not a direct global-admin takeover; but cross-app builder is effectively a tenant-wide app/data-plane compromise. No real secrets were accessed during testing; the lab used canary-only data.",
"id": "GHSA-j9fc-w3mr-x6mv",
"modified": "2026-08-12T18:57:18Z",
"published": "2026-07-24T21:12:40Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/Budibase/budibase/security/advisories/GHSA-j9fc-w3mr-x6mv"
},
{
"type": "PACKAGE",
"url": "https://github.com/Budibase/budibase"
},
{
"type": "WEB",
"url": "https://github.com/Budibase/budibase/releases/tag/3.40.0"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": " Budibase: Privilege escalation via public role assignment API missing app-level authorization"
}
Drag nodes to arrange them. Double-click a related vulnerability to open it, or a source to visit its link. The Physics button on the left tunes the layout and the simulation; the legend at the bottom left hides or shows nodes by kind and by sighting type.
Sightings
| Author | Source | Type | Date | Other |
|---|
Nomenclature
- Seen: The vulnerability was mentioned, discussed, or observed by the user.
- Confirmed: The vulnerability has been validated from an analyst's perspective.
- Published Proof of Concept: A public proof of concept is available for this vulnerability.
- Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
- Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
- Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
- Not confirmed: The user expressed doubt about the validity of the vulnerability.
- Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
Vulnerability timeline
Sightings and recorded dates, grouped by day · UTC
How durations are calculated
Following the VTEM framework (Black Hat USA 2015), time to patch starts when the vendor or a coordinator was informed, total time to patch at discovery, and disclosure to patch at public disclosure; vendor response time runs from the report to its acknowledgement. Time to exploit and time unpatched start at the earliest discovery or report, falling back to public disclosure. Unpatched time ends at the first recorded patch, or the calculation time when no patch is recorded. Exploitation to patch shows the interval between those two events; a published proof of concept is a public exploit, not exploitation.
Each result shows its source dates. A negative interval means the end event came first; when date-only inputs leave the order open, the result says so. Date-only inputs are approximate. The earliest evidence wins: KEV report dates and sightings are labelled as such. Records describing another vulnerability are listed but never calculated from, and a PoC alone does not establish exploitation. Missing patch evidence does not prove that no patch exists. Filters below change the event list only.
Loading timeline…
The approach is described in our paper Mapping CVEs to MITRE ATT&CK Techniques: A Curated Gold-Set Classifier and the Limits of LLM-Assisted Label Expansion.
Browse all ATT&CK techniques and the vulnerabilities related to each.
Related by attack behaviour
Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.