packet-autopsy.com

Case 018 · Incident: Feb–Jul 2026

ForgeRock car hack: The password "admin" unlocked cars at two automakers

In early 2026, researchers Shubham Shah and Sam Curry reverse-engineered the Android apps of two unnamed automakers and found that both signed customers in through ForgeRock's identity platform (OpenAM/OpenIDM, now PingIDM). On the US/production servers they bypassed a WAF via path-confusion, reached an Apache Felix administrator console still protected by the documented default admin:admin credentials, achieved remote code execution via the OpenIDM Groovy scheduler, and could create a session for any customer account—giving the same remote vehicle control as the owner's app (unlock, engine start/stop, location). A companion write-up by Shah detailed a separate Australian chain starting from a license plate, exploiting hardcoded admin credentials left in the app's JavaScript and a skippable MFA step.

Watch on YouTube

The Car That Unlocked Itself

During a test in the United States, a test car unlocked its doors and started its engine. Its owner was standing right beside it, and nobody had touched the car.[1] Moments later the same unseen hand switched the engine off and locked the doors again.[1] Those commands came through the owner's account. They were sent with the account's own built-in remote-control features.[1][2]

They held a login, and that login carried everything the owner's own app could do.[2] Its location and live telemetry, its door locks, and remote engine start and stop.[2]

Rewind: One Front Door

To see how a login became a car key, go back to early 2026.[2] Two researchers, Shubham Shah and Sam Curry, downloaded both automakers' Android apps and took them apart.[2] It was not their first time; they had already reached cars from Honda, Kia and Subaru, among others.[1]

Tracing each app's login, they found both led back to the same identity platform, ForgeRock.[2] It runs in two parts: one signs customers in, the other, OpenIDM, holds the customer records.[1][3] Both brands ran the same platform, and both had left the same door open. The researchers say the mistake was in how the brands ran it, not a new bug in the product.[1][2][3]

OpenIDM ships with a built-in administrator console, run by a tool called Apache Felix.[1] ForgeRock's own guide for hardening a live system recommends removing or protecting that console. It also notes that the console handles logins separately from the rest of OpenIDM.[2]

admin:admin

On the live servers it sat behind a web application firewall, a filter meant to block sensitive pages.[1] The researchers sent a request the firewall and the server read two different ways, and the filter waved it through as a harmless file.[1] Behind it, the console still used the documented default: the word admin for both the username and the password.[1]

It worked.[2] The console listed 361 software components, 343 of them active, and opened its configuration manager.[4] From there they abused the scheduler's support for inline scripts to run their own code inside the identity process.[4]

That code ran inside the live identity process itself. This was the process that read and changed the customer records behind both automakers' logins.[5] The chain was reproduced on one live server behind each automaker's login platform.[2] In the researchers' own words, this was remote code execution on production servers.[2]

Any Owner's Keys

Now any customer account could be found by name, email, phone number or internal ID, and opened.[5] The exposed accounts included vehicle identification numbers, location and mileage. They also showed battery or fuel range, tyre pressure and subscription details.[5] Using the account's own commands, they unlocked, started, stopped and re-locked a real test car.[1]

This was a controlled test, with the owner's permission, but the access was entirely real.[1] Changing the account's email and password did not even log the real owner out.[1] And the account's alerts could be quietly redirected to the attacker's phone.[1]

The researchers say the affected systems served customers in several regions, though the full chain was tested only on US cars.[2] Those tests ran in February 2026, on the researchers' own and consenting owners' vehicles.[1][6] Two stranger things come next: a chain that starts from a licence plate, and why this flaw has no official number.

A Plate and Six Hours

In July, the pair turned to an Australian connected-car system.[7] There, a licence plate alone yields the car's VIN through state registration lookup sites.[7] This automaker had left administrator credentials inside the Android app handed to every customer.[8]

A separate internet-facing admin tool returned a page titled TEST HARNESS, and many of its interfaces needed nothing more than keys hardcoded in its JavaScript. The admin credentials inside the Android app had a problem of their own.[7][8] Those credentials were meant for staging, a separate test copy of the system. Given the sloppy security they had already seen, the researchers expected the credentials to work on the live production system too.[7] One firewall-exposed endpoint handed back password and PIN hashes, push tokens, phone numbers and full names.[8]

Put together, the Australian path ran from a number plate, to a VIN, to personal account details. Along the way, the servers barely checked who was asking.[7][8] And the multi-factor step, the extra code meant to stop stolen passwords, could be skipped entirely.[7] Shah's post adds one detail about how they found a working firewall bypass. They let an AI model, GPT 5.6 Sol Ultra, study the OpenIDM source code for about six hours.[7]

Per the researchers, the account passwords were hashed with PBKDF2-SHA512, a deliberately slow algorithm.[7] Shah says the two write-ups together describe sending unlock and engine-start commands to cars in Australia, the United States and Canada. Whether his Australian carmaker is one of Curry's two is not clear, and only US accounts and vehicles were fully tested.[2][7]

Two Hours, No CVE

Curry's two automakers fixed it about two hours after the report, and rotated the affected credentials.[1][2] Neither write-up reports any malicious use, or says whether the researchers checked for it.[3] According to iTnews, the Australian carmaker offered no bounty and no credit, but still fixed it within hours.[8]

There is no CVE and no vendor advisory, because the fault was in how the brands ran ForgeRock, not a product bug.[1] The work went public on November 7, 2026.[6] At the researchers' request, the carmakers were not named.[6]

Autopsy Verdict

The chain needed both steps. The firewall trick got the request through to the console, and the documented default password, admin and admin, opened it.[1][2] The vendor's own hardening guide had advised removing or protecting that console. On these live servers it was still there, and it still used its documented default password.[1][2] Unproven: no malicious use is reported, and no one says whether prior abuse was ever checked.[3]

Unknown: which automakers these were; only Tom's Hardware guessed one might be Toyota.[6][9] With no CVE to flag it, the clearest warning was the one already in the vendor's hardening guide.[1][2]

What to do

  • Remove or lock down built-in admin consoles, and change every default password before a system goes live.
  • Do not trust a firewall to hide an admin page by its path; the edge and the server can read one request differently.
  • Never bundle passwords or keys inside a mobile app, and keep staging and production secrets apart.
  • If you build account systems, end every active session and alert the old contacts whenever an account's email or password changes. Check multi-factor codes on the server, so no app can skip the step.

Open questions

  • The identities of the two automakers were withheld; it is unconfirmed whether Automaker #2 is Toyota (only speculated by Tom's Hardware).
  • It is unclear whether the two unnamed US/Canada automakers are the same company as the Australian two-brand carmaker in Shah's companion post.
  • Neither write-up states whether anyone else discovered or exploited the path before it was closed, nor whether the researchers checked for prior malicious use.
  • Shah's Australian post does not say when the Australian findings were reported or whether they were fixed.
  • Whether European connected-car environments were affected remains open; Shah notes US, Canadian and European environments were better protected but makes no claim about Europe being hit.
  • Ping Identity did not respond to a request for comment by publication time.

Sources

  1. [1]The Hacker News — Default 'admin:admin' Server Password Helps Researchers Hack Two Automakers, Unlock and Start Cars
  2. [2]Sam Curry — Hacking Two Automakers: ForgeRock Path Confusion, RCE, and Remote Vehicle Access
  3. [3]Infosecurity Magazine — Default Admin Password Let Researchers Unlock and Start Cars
  4. [4]GBHackers — Hackers Exploit ForgeRock WAF Bypass and RCE to Remotely Unlock and Start Cars From Two Automakers
  5. [5]Cybernews
  6. [6]Pasquale Pillitteri
  7. [7]Shubham Shah (shubs.io) — unlocking any car from X brand in most of the world
  8. [8]iTnews
  9. [9]Tom's Hardware — All net-connected vehicles from two major automakers allowed hackers to remotely unlock doors and start the engine