Google Passkey Weaknesses Expose Keys as AI Agents Automate Cyberattacks

Passkeys replace reusable passwords with public-private key pairs, reducing exposure to phishing and credential reuse. Google Password Manager extends that model by synchronizing encrypted passkeys across devices, but the convenience shifts more security responsibility to endpoints and cloud authentication infrastructure. Separately, connecting generative AI models to terminals and agent frameworks can automate reconnaissance, vulnerability selection and exploitation attempts, potentially compressing cyberattack timelines and expanding their scale.

On Aug. 3, Palo Alto Networks’ Unit 42 detailed three post-compromise techniques — Pass-ta-key, Silver Pass-ta-key and Golden Pass-ta-key — against Google Password Manager on Windows. Malware already running on a device could sign in without a PIN, biometric check or visible prompt; the most severe method extracts the 32-byte Security Domain Secret from Chrome memory to decrypt synced private keys. In separate research published July 31, Unit 42 said a China-based actor used DeepSeek with Hermes Agent to attempt autonomous attacks on roughly 460 targets. Researchers confirmed three Citrix NetScaler compromises but attributed those successful intrusions to manual activity.