1Password gives Claude controlled access to stored logins

1Password has launched 1Password for Claude, a browser integration that lets Anthropic PBC’s Claude use stored credentials to complete online tasks without showing the password or multifactor code to the AI model. The release targets a basic security gap in AI agents: getting an assistant into a logged-in service without pasting secrets into a chat window or stopping the workflow for every manual sign-in.

The company did not announce new pricing for the Claude integration. It said the feature is available now to 1Password users on Mac across business, family and individual plans.

According to 1Password, credentials are sent to the destination site through a secure channel controlled by 1Password. The company says the password and any one-time multifactor code remain outside Claude’s context, memory and Anthropic’s systems. Claude asks for access to credentials needed for a specific task, and the user approves or rejects each request through a biometric prompt.

Access is limited by session and by item. 1Password said approvals apply only to selected credentials for the current task and do not create ongoing access for later sessions. That design is meant to avoid the common pattern of giving an agent a reusable secret, then relying on policy or prompt discipline to keep it contained.

Agent access gets a narrower permission model

The Claude launch runs on what 1Password calls a zero-exposure security framework. Alongside the integration, the company is rolling out Agentic Mode to all users. When a compatible agent takes control of the browser, 1Password says the user’s vault is locked down and only credentials explicitly allowed for that task remain available.

Agentic Mode activates automatically while an agent is working. Users can see the mode in the 1Password browser extension and can turn it off. The company is positioning the feature as a control layer for browser-based agents, starting with Claude and extending later to other agents. It did not name additional agent partners or give a timeline.

1Password Chief Technology Officer Nancy Wang said in the company’s announcement that agent security needs a model built for software acting on a user’s behalf, rather than one that hands secrets to the agent. Her framing reflects a broader enterprise concern as AI assistants move from summarizing and drafting into taking actions across web apps.

The release also includes two workflow controls. 1Password said it can broker credential access across multiple sites within a single task, allowing Claude to complete multistep processes without repeatedly asking for logins. The company also said it scans pages after each autofill and removes filled values if a form submission fails, reducing the chance that secrets remain visible in the browser.

The first version is limited. 1Password for Claude requires the 1Password desktop app and browser extensions, as well as the Claude desktop app and browser extensions. Support for payment cards and personal details, including names and addresses, is planned after launch, according to the company.

1Password said its enterprise vault contains more than 1.5 billion credentials and secrets. The company says it is used by more than 1 million developers and 180,000 businesses, including Figma, GitHub, MongoDB, Notion, Perplexity AI, Salesforce and Stripe.

This story draws on original reporting from SiliconANGLE.