Anthropic has expanded the capabilities of its AI chatbot Claude through a partnership with password manager 1Password, enabling the assistant to securely log into websites and complete real-world tasks on behalf of users.
The new integration allows Claude to perform actions such as booking travel, managing online accounts and shopping without directly accessing or storing users' passwords.
According to 1Password, Claude can retrieve login credentials, one-time passcodes and other authentication details when users authorize a task. However, the credentials never enter the AI model or its memory.
"Claude can complete browser tasks that require logins and one-time passcodes, but the credentials never enter the model or its memory," 1Password said.
The company added that access is limited to the specific task being performed and automatically ends once the task is completed. After autofilling credentials, 1Password also verifies that sensitive information has not been exposed on the webpage.
Users will still need to approve sensitive actions using biometric authentication, such as a fingerprint or facial recognition, through their smartphones.
Nancy Wang, Chief Technology Officer at 1Password, said the company has developed a new security model specifically designed for AI agents to ensure user credentials remain protected.
The feature marks another step in the rise of "agentic AI," where artificial intelligence systems move beyond answering questions to carrying out tasks independently across browsers and applications.
While the technology promises greater convenience, cybersecurity experts have raised concerns over the growing autonomy of AI agents.
One widely reported incident in April saw a Claude-powered AI agent accidentally delete an entire production database in just nine seconds, highlighting the risks of allowing AI systems to perform sensitive operations.
A recent survey by Economist Enterprise involving more than 800 businesses across nine countries found that 98% of respondents had experienced disruptive AI-related incidents, while many organizations lacked full visibility into how their AI agents operated.
Researchers have also warned that cybercriminals are increasingly using AI agents to automate ransomware attacks and other malicious campaigns, lowering the technical barriers for hackers.
Despite the concerns, Anthropic and 1Password say the new safeguards are designed to ensure users retain control while enabling Claude to handle increasingly complex digital tasks securely.
