Google Password Manager Attacks: The "Golden Pass-ta-key" Risk

Google Password Manager Attacks: The "Golden Pass-ta-key" Risk

Just saw the Unit 42 report regarding the new attack vectors targeting Chrome's Google Password Manager (GPM). It is deeply concerning because it fundamentally undermines the assumption that passkeys are inherently phishing-resistant and require user presence (biometrics/PIN).

The research details three distinct paths—Pass-ta-key, Silver Pass-ta-key, and Golden Pass-ta-key. The most critical, Golden Pass-ta-key, allows malware running as a standard user to bypass the user verification screen entirely. By extracting the GPM cloud authenticator's master key from the Windows Credential Manager, attackers can generate valid passkey assertions locally without any interaction on the victim's screen.

This attack relies heavily on how Windows stores these credentials using the Data Protection API (DPAPI). If the malware has the user's context, it can often unwrap these DPAPI blobs to decrypt the master key.

Detection Opportunities

Since this requires reading the GPM database and registry keys, we should look for anomalous access patterns. Specifically, non-browser processes touching the GPM registry hive or unusual calls to DPAPI.

Here is a PowerShell snippet to audit recent process access to the GPM Registry keys (assuming Object Access auditing is enabled):

Get-WinEvent -LogName Security -MaxEvents 1000 | Where-Object {
    $_.Message -match "Object Name.*Google\\Password Manager" -and
    $_.Message -match "Accesses.*KEY_READ"
} | Select-Object TimeCreated, Id, Message | Format-List


EDR solutions should also trigger on unauthorized processes attempting `CryptUnprotectData` calls on memory regions associated with Chrome if they aren't signed by Google.

**Discussion:** Passkeys were supposed to solve the "shared secret" problem, but this proves that the endpoint security model is still the bottleneck. For those of you managing Windows fleets, are you allowing Google Password Manager, or are you enforcing hardware-bound FIDO2 keys only?