GHSA-VWRP-X96C-MHWQ
Vulnerability from github – Published: 2026-05-07 04:07 – Updated: 2026-05-14 20:36Summary
vm2's bridge exposes mutable proxies for real host-realm intrinsic prototypes and then forwards sandbox writes into the underlying host objects with otherReflectSet() and otherReflectDefineProperty(), which lets attacker-controlled JavaScript running in a default VM or inherited NodeVM mutate shared host Object.prototype, Array.prototype, and Function.prototype from inside the sandbox.
Details
BaseHandler.apply() unwraps sandbox-controlled receivers and arguments with otherFromThis() / otherFromThisArguments() and then directly invokes the real host function with ret = otherReflectApply(object, context, args), so any default-exposed host function that can surface a prototype getter becomes a prototype-walking primitive (lib/bridge.js:665-676). BaseHandler.get() special-cases proto and returns the host-side descriptor or proxy target prototype, which is enough for the attacker to reuse the host lookupGetter('proto') accessor repeatedly until the walk lands on host Object.prototype, Array.prototype, or Function.prototype (lib/bridge.js:590-616). Once the attacker has a proxy to a host intrinsic prototype, BaseHandler.set() performs value = otherFromThis(value); return otherReflectSet(object, key, value) === true;, which writes attacker-controlled data directly into the shared host object instead of keeping the mutation sandbox-local; BaseHandler.defineProperty() repeats the same design at otherReflectDefineProperty(object, prop, otherDesc) for descriptor-based writes (lib/bridge.js:641-649, lib/bridge.js:753-774). Existing validation does not stop the attack because the constructor filter only blocks one dangerous-property access pattern, setPrototypeOf() only blocks prototype replacement rather than ordinary property assignment, and containsDangerousConstructor() only protects one later re-unwrapping path instead of the initial host-prototype write sink (lib/bridge.js:494-530, lib/bridge.js:595-610, lib/bridge.js:660-662).
PoC
Run the following code snippet and observe that the value of vm2EscapeMarker is polluted:
const { VM } = require('vm2');
const vm = new VM();
vm.run(`
const g = ({}).__lookupGetter__;
const a = Buffer.apply;
const p = a.apply(g, [Buffer, ['__proto__']]);
const hostObjectProto = p.call(p.call(p.call(p.call(Buffer.of()))));
hostObjectProto.vm2EscapeMarker = 'polluted-object-prototype';
`);
console.log({}.vm2EscapeMarker)
Impact
Sandbox escape and prototype pollution.
{
"affected": [
{
"database_specific": {
"last_known_affected_version_range": "\u003c= 3.10.5"
},
"package": {
"ecosystem": "npm",
"name": "vm2"
},
"ranges": [
{
"events": [
{
"introduced": "3.9.6"
},
{
"fixed": "3.11.0"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-44005"
],
"database_specific": {
"cwe_ids": [
"CWE-1321",
"CWE-94"
],
"github_reviewed": true,
"github_reviewed_at": "2026-05-07T04:07:05Z",
"nvd_published_at": "2026-05-13T18:16:17Z",
"severity": "CRITICAL"
},
"details": "### Summary\nvm2\u0027s bridge exposes mutable proxies for real host-realm intrinsic prototypes and then forwards sandbox writes into the underlying host objects with otherReflectSet() and otherReflectDefineProperty(), which lets attacker-controlled JavaScript running in a default VM or inherited NodeVM mutate shared host Object.prototype, Array.prototype, and Function.prototype from inside the sandbox.\n\n### Details\nBaseHandler.apply() unwraps sandbox-controlled receivers and arguments with otherFromThis() / otherFromThisArguments() and then directly invokes the real host function with ret = otherReflectApply(object, context, args), so any default-exposed host function that can surface a prototype getter becomes a prototype-walking primitive ([lib/bridge.js:665-676](https://github.com/patriksimek/vm2/blob/408fc855f1cc1bbc2985b029465ee0e732ada433/lib/bridge.js#L665-L676)). BaseHandler.get() special-cases __proto__ and returns the host-side descriptor or proxy target prototype, which is enough for the attacker to reuse the host __lookupGetter__(\u0027__proto__\u0027) accessor repeatedly until the walk lands on host Object.prototype, Array.prototype, or Function.prototype ([lib/bridge.js:590-616](https://github.com/patriksimek/vm2/blob/408fc855f1cc1bbc2985b029465ee0e732ada433/lib/bridge.js#L590-L616)). Once the attacker has a proxy to a host intrinsic prototype, BaseHandler.set() performs value = otherFromThis(value); return otherReflectSet(object, key, value) === true;, which writes attacker-controlled data directly into the shared host object instead of keeping the mutation sandbox-local; BaseHandler.defineProperty() repeats the same design at otherReflectDefineProperty(object, prop, otherDesc) for descriptor-based writes ([lib/bridge.js:641-649](https://github.com/patriksimek/vm2/blob/408fc855f1cc1bbc2985b029465ee0e732ada433/lib/bridge.js#L641-L649), [lib/bridge.js:753-774](https://github.com/patriksimek/vm2/blob/408fc855f1cc1bbc2985b029465ee0e732ada433/lib/bridge.js#L753-L774)). Existing validation does not stop the attack because the constructor filter only blocks one dangerous-property access pattern, setPrototypeOf() only blocks prototype replacement rather than ordinary property assignment, and containsDangerousConstructor() only protects one later re-unwrapping path instead of the initial host-prototype write sink ([lib/bridge.js:494-530](https://github.com/patriksimek/vm2/blob/408fc855f1cc1bbc2985b029465ee0e732ada433/lib/bridge.js#L494-L530), [lib/bridge.js:595-610](https://github.com/patriksimek/vm2/blob/408fc855f1cc1bbc2985b029465ee0e732ada433/lib/bridge.js#L595-L610), [lib/bridge.js:660-662](https://github.com/patriksimek/vm2/blob/408fc855f1cc1bbc2985b029465ee0e732ada433/lib/bridge.js#L660-L662)).\n\n### PoC\nRun the following code snippet and observe that the value of vm2EscapeMarker is polluted:\n```\nconst { VM } = require(\u0027vm2\u0027);\nconst vm = new VM();\nvm.run(`\n const g = ({}).__lookupGetter__;\n const a = Buffer.apply;\n const p = a.apply(g, [Buffer, [\u0027__proto__\u0027]]);\n const hostObjectProto = p.call(p.call(p.call(p.call(Buffer.of()))));\n hostObjectProto.vm2EscapeMarker = \u0027polluted-object-prototype\u0027;\n`);\nconsole.log({}.vm2EscapeMarker)\n```\n\n### Impact\nSandbox escape and prototype pollution.",
"id": "GHSA-vwrp-x96c-mhwq",
"modified": "2026-05-14T20:36:31Z",
"published": "2026-05-07T04:07:05Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/patriksimek/vm2/security/advisories/GHSA-vwrp-x96c-mhwq"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-44005"
},
{
"type": "PACKAGE",
"url": "https://github.com/patriksimek/vm2"
},
{
"type": "WEB",
"url": "https://github.com/patriksimek/vm2/releases/tag/v3.11.0"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:N/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "vm2: Mutable Proxies for Host Intrinsic Prototypes Allows Sandbox Escape"
}
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.