1Password might just have a solution to the password security problem posed by AI agents (www.techradar.com)

1Password has unveiled Secure Agentic Autofill, a system designed to stop AI agents from seeing or copying credentials while still allowing them to authenticate as part of automated workflows. Instead of handing secrets to an agent or embedding them in prompts, credentials remain in the 1Password browser extension; an agent requests a login, a human approves the request, and the extension fills the credentials directly so the agent never accesses the secret. The solution uses an encrypted device-to-extension channel built on the Noise Framework and 1Password says partners such as Browserbase will integrate the feature into browser automation UIs with a zero-access model for the agent and the automation service. This matters because agentic, often headless, browsers are increasingly used to automate tasks but can inadvertently leak credentials into LLM prompts, logs, or systems outside identity controls—creating large attack surfaces and compliance blind spots. Secure Agentic Autofill aims to reduce that risk by keeping secrets out of agents, enabling centralized credential management and revocation, and adding human approval for use. Technically, it’s a client-side, encrypted autofill proxy that preserves automation efficiency while improving auditability and limiting credential sprawl across agents and infrastructure.