Pass-ta-key attack exposes Windows passkey architecture, not a new vulnerability

A security researcher from Palo Alto Networks has demonstrated an attack dubbed 'Pass-ta-key' that extracts passkeys from the Google Password Manager on Windows devices infected with malware. Arie Olshtein, the researcher behind the demonstration, outlined how the attack exploits the architectural differences in how Windows handles passkey storage compared to other operating systems. The findings have generated discussion among security professionals regarding the safety of the new authentication paradigm, though experts note the technique is not novel.

Contrary to common belief, the FIDO 2 specifications managed by the FIDO Alliance do not mandate that passkeys be stored in Trusted Platform Modules or dedicated hardware enclaves. While Microsoft offers this option for enterprise users, most platforms and third-party software store passkeys locally on the device to facilitate syncing. This design shift occurred after developers realised that requiring hardware-only storage hindered widespread adoption, as syncing keys across devices would otherwise be impossible without recreating each one individually.

Windows passkey applications, including Google Password Manager, 1Password, and Dashlane, typically store keys in end-to-end encrypted blobs on the server to mitigate the risk of local malware harvesting. The device uses the TPM only for user or device keys to unlock these blobs. When a user attempts to log in, the device presents a key to the server, which then provides an authentication assertion signed with the private key stored in encrypted form on the server. This approach contrasts with other platforms where the local device presents the assertion directly.

The 'Pass-ta-key' attack variants include a scenario where the infected Windows machine masquerades as an iPhone to trigger the sync capability in Google Password Manager. This allows the transfer of all stored passkeys to the compromised device. Olshtein described the attack as targeting a novel surface in the passkey ecosystem, but the underlying risk mirrors existing threats where malware on a logged-in device can access sensitive data.

The demonstration highlights that Windows apps generally run with the privileges of the user, unlike the stricter sandboxing on macOS, iOS, and Android. While Windows provides some sandboxing protections, it does not prevent unsandboxed apps such as malware from accessing the data of sandboxed apps. Consequently, malware on a compromised Windows machine can harvest stored credentials just as it might harvest passwords from other credential management applications.

Ultimately, the attack serves as a reminder that passkeys are designed to eliminate shared secrets vulnerable to phishing or server breaches, not to withstand physical attacks or malware infections on the device. Once a device is compromised while logged into an account, the data stored there is accessible to the attacker. This risk is inherent to computing security and applies to any data requiring authentication for access, regardless of whether the system uses passkeys or traditional passwords.