Password management platform 1Password and artificial intelligence company Anthropic have announced a new integration that enables Claude AI agents to securely access approved credentials without exposing passwords, API keys or other sensitive secrets to the underlying AI model. The collaboration is designed to address one of the key enterprise challenges in deploying autonomous AI agents: enabling them to perform real-world tasks while maintaining strict security controls.
The integration combines 1Password's Extended Access Management (XAM) platform with Anthropic's Model Context Protocol (MCP), allowing AI agents powered by Claude to retrieve credentials only after explicit user approval. Rather than revealing the actual secret to the AI model, the system provides a secure session that enables the agent to complete authorised tasks without directly accessing or storing sensitive information.
According to the companies, the approach introduces a security layer that keeps credentials protected while still allowing AI agents to interact with enterprise systems. This addresses growing concerns around the use of AI assistants for operational workflows that require authentication across applications, cloud environments and enterprise platforms.
The new capability is intended for enterprise use cases where AI agents assist employees with activities such as software development, IT administration, DevOps operations and business automation. Developers can authorise Claude to access specific systems or services through 1Password without manually copying passwords or API keys into AI conversations.
The companies said the integration is built around a "human in the loop" approval model. Every request for credential access requires user authorisation before an AI agent can proceed. Once approved, the credentials remain isolated within a secure execution environment, ensuring that neither Claude nor the underlying large language model can view, retain or reproduce the sensitive information.
The announcement reflects growing enterprise demand for secure AI deployment as organisations increasingly integrate generative AI into business operations. While AI agents are becoming capable of performing more complex actions across enterprise software, concerns around credential management, data privacy and regulatory compliance have remained significant barriers to broader adoption.
Security experts have highlighted that exposing passwords or API keys directly to AI models creates potential risks, including accidental leakage, prompt injection attacks or unauthorised access. By separating credential management from the AI model itself, the new integration aims to reduce these risks while enabling more practical enterprise automation.
The partnership also builds on Anthropic's Model Context Protocol, an open standard introduced to connect AI assistants with external tools, applications and data sources. The protocol has gained traction across the AI ecosystem as companies look for standardised methods to integrate AI agents with enterprise infrastructure while maintaining governance and security.
For 1Password, the integration extends its strategy beyond traditional password management into enterprise identity and access security. The company has increasingly focused on securing human users, applications and now AI agents as organisations adopt autonomous workflows powered by generative AI.
Industry analysts believe secure identity management will become a foundational requirement as AI agents evolve from conversational assistants into software capable of executing tasks across multiple enterprise environments. Technologies that combine identity verification, approval workflows and secure credential handling are expected to play a growing role in enterprise AI deployments.
The collaboration between 1Password and Anthropic signals a broader industry shift towards building security directly into AI infrastructure rather than treating it as an additional layer. As businesses continue adopting AI agents across critical operations, secure access management is likely to become an essential component of enterprise AI strategies.