GHSA-HF59-9XWC-2G4P
Vulnerability from github – Published: 2026-11-16 15:02 – Updated: 2026-11-16 15:02Summary
The fix for CVE-2026-46342 (GHSA-g8wj-3cr3-6w7v) binds each /__nuxt_island/<Name>_<hash>.json path to its props by hashing the (name, props, context, source) tuple. That hash was computed over ohash's serialization, which does not escape quotes inside strings or delimiters inside keys, so two different prop objects can produce the same hash. A request with crafted props can therefore pass validation for another island's path and render different content at that path.
Impact
For example, these two prop objects serialize identically and share one island path:
{ html: '', note: '<img src=x onerror="alert(1)">' }
{ html: '\',note:\'<img src=x onerror="alert(1)">' }
This reopens the scenarios described in GHSA-g8wj-3cr3-6w7v:
nuxt generate/ prerendering. Nuxt caches prerendered island responses by path. If the crawler reaches a link to the colliding URL (for example in user-submitted content included in the build) before it renders the page that uses the legitimate props, the colliding response is written into the generated page.- Path-keyed shared caches. A CDN or reverse proxy that caches
/__nuxt_island/*by path only, ignoring the query, can be primed with the colliding response.
An attacker has to construct props that collide with a specific legitimate tuple, which in practice means at least part of that tuple is a string they control (for example a stored comment or profile field passed as a prop). Where a prop reaches an unsafe HTML sink in the island (v-html, innerHTML, a renderer treating a prop as HTML), the substituted response can execute script in the site's origin. Without such a sink, the impact is content substitution.
Applications are not affected if they do not use server components or <NuxtIsland>, or if no island receives attacker-influenced string props. Server-rendered responses that are not cached by path alone are also unaffected.
Patches
Fixed in nuxt@4.6.4 and @nuxt/nitro-server@4.6.4. The island hash is now computed over a canonical JSON encoding of the tuple, in which strings and keys are escaped, so different props can no longer produce the same hash.
Island hashes change as a result. Clients that build island URLs themselves should use the exported getIslandHash rather than hashing with ohash directly.
Nuxt 3 has reached end of life and will not receive a fix. Upgrade to Nuxt 4.
Workarounds
- Ensure any intermediary cache keys
/__nuxt_island/*on the full query string, not on the path alone. - Treat island props as untrusted input and do not pass them to
v-htmlor other HTML sinks without sanitisation. - When prerendering, avoid including user-submitted links to
/__nuxt_island/in crawled pages, for example by excluding that prefix withnitro.prerender.ignore:
export default defineNuxtConfig({
nitro: {
prerender: {
ignore: [route => route.startsWith('/__nuxt_island/') && route.includes('?')],
},
},
})
Credits
Thanks to rat5ak for reporting this issue.
{
"affected": [
{
"package": {
"ecosystem": "npm",
"name": "nuxt"
},
"ranges": [
{
"events": [
{
"introduced": "4.4.6"
},
{
"fixed": "4.6.4"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"database_specific": {
"last_known_affected_version_range": "\u003c 4.0.0"
},
"package": {
"ecosystem": "npm",
"name": "nuxt"
},
"ranges": [
{
"events": [
{
"introduced": "3.21.6"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"package": {
"ecosystem": "npm",
"name": "@nuxt/nitro-server"
},
"ranges": [
{
"events": [
{
"introduced": "4.4.6"
},
{
"fixed": "4.6.4"
}
],
"type": "ECOSYSTEM"
}
]
},
{
"database_specific": {
"last_known_affected_version_range": "\u003c 4.0.0"
},
"package": {
"ecosystem": "npm",
"name": "@nuxt/nitro-server"
},
"ranges": [
{
"events": [
{
"introduced": "3.21.6"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-128076"
],
"database_specific": {
"cwe_ids": [
"CWE-116",
"CWE-349"
],
"github_reviewed": true,
"github_reviewed_at": "2026-11-16T15:02:22Z",
"nvd_published_at": "2026-11-13T18:17:37Z",
"severity": "LOW"
},
"details": "### Summary\n\nThe fix for CVE-2026-46342 (GHSA-g8wj-3cr3-6w7v) binds each `/__nuxt_island/\u003cName\u003e_\u003chash\u003e.json` path to its props by hashing the `(name, props, context, source)` tuple. That hash was computed over `ohash`\u0027s serialization, which does not escape quotes inside strings or delimiters inside keys, so two different prop objects can produce the same hash. A request with crafted props can therefore pass validation for another island\u0027s path and render different content at that path.\n\n### Impact\n\nFor example, these two prop objects serialize identically and share one island path:\n\n```js\n{ html: \u0027\u0027, note: \u0027\u003cimg src=x onerror=\"alert(1)\"\u003e\u0027 }\n{ html: \u0027\\\u0027,note:\\\u0027\u003cimg src=x onerror=\"alert(1)\"\u003e\u0027 }\n```\n\nThis reopens the scenarios described in GHSA-g8wj-3cr3-6w7v:\n\n- **`nuxt generate` / prerendering.** Nuxt caches prerendered island responses by path. If the crawler reaches a link to the colliding URL (for example in user-submitted content included in the build) before it renders the page that uses the legitimate props, the colliding response is written into the generated page.\n- **Path-keyed shared caches.** A CDN or reverse proxy that caches `/__nuxt_island/*` by path only, ignoring the query, can be primed with the colliding response.\n\nAn attacker has to construct props that collide with a specific legitimate tuple, which in practice means at least part of that tuple is a string they control (for example a stored comment or profile field passed as a prop). Where a prop reaches an unsafe HTML sink in the island (`v-html`, `innerHTML`, a renderer treating a prop as HTML), the substituted response can execute script in the site\u0027s origin. Without such a sink, the impact is content substitution.\n\nApplications are not affected if they do not use server components or `\u003cNuxtIsland\u003e`, or if no island receives attacker-influenced string props. Server-rendered responses that are not cached by path alone are also unaffected.\n\n### Patches\n\nFixed in `nuxt@4.6.4` and `@nuxt/nitro-server@4.6.4`. The island hash is now computed over a canonical JSON encoding of the tuple, in which strings and keys are escaped, so different props can no longer produce the same hash.\n\nIsland hashes change as a result. Clients that build island URLs themselves should use the exported `getIslandHash` rather than hashing with `ohash` directly.\n\nNuxt 3 has reached end of life and will not receive a fix. Upgrade to Nuxt 4.\n\n### Workarounds\n\n- Ensure any intermediary cache keys `/__nuxt_island/*` on the full query string, not on the path alone.\n- Treat island props as untrusted input and do not pass them to `v-html` or other HTML sinks without sanitisation.\n- When prerendering, avoid including user-submitted links to `/__nuxt_island/` in crawled pages, for example by excluding that prefix with `nitro.prerender.ignore`:\n\n```ts\nexport default defineNuxtConfig({\n nitro: {\n prerender: {\n ignore: [route =\u003e route.startsWith(\u0027/__nuxt_island/\u0027) \u0026\u0026 route.includes(\u0027?\u0027)],\n },\n },\n})\n```\n\n### Credits\n\nThanks to rat5ak for reporting this issue.",
"id": "GHSA-hf59-9xwc-2g4p",
"modified": "2026-11-16T15:02:22Z",
"published": "2026-11-16T15:02:22Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/nuxt/nuxt/security/advisories/GHSA-hf59-9xwc-2g4p"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-128076"
},
{
"type": "WEB",
"url": "https://github.com/nuxt/nuxt/commit/6f1b93c643355688bc9e2d7434723f7fd3f25092"
},
{
"type": "PACKAGE",
"url": "https://github.com/nuxt/nuxt"
},
{
"type": "WEB",
"url": "https://github.com/nuxt/nuxt/releases/tag/v4.6.4"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:H/PR:N/UI:R/S:U/C:N/I:L/A:N",
"type": "CVSS_V3"
}
],
"summary": "Nuxt: Colliding island props can share one `/__nuxt_island` path (incomplete fix for CVE-2026-46342)"
}
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.
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.